Oh, yes - this is a good point, as an argument why selection against roadkill behavior might not occur.
That’s a big factor. Cars have been around for only a little more than a century. That’s an eyeblink as far as evolution is concerned, especially since with creatures far more complex than, say, bacteria.
It’s still possible that you might see a noticeable change in behavior in a few generations if you are aggressively “culling” in significant numbers. But there are so many unanswered questions about the OP. We have no data on what’s really happening.
Actually, here in Shanghai, I’ve seen a cat wait at a pedestrian crossing, and then cross, calmly, when people started crossing. And I’ve seen a pack of feral dogs waiting for an opening in traffic big enough for them all to cross together.
Both things were adorable to see, but are unlikely to be the result of natural selection. It’s likely just trial and error on the part of individuals (when cars are slower, it’s possible for animals to have near-misses and readjust their behaviour accordingly). But still, I think it shows the situation is a bit more complex than implied in the OP.
I don’t think this is right. Any single factor may only account for a couple of percent of deaths, but resilience to any one of these factors will have a strong selection pressure regardless.
I think it’s more likely just that it’s such an unusual threat, that’s suddenly appeared, that it will take a while for the unlikely set of mutations to occur. But if hypothetically this situation were to remain the same for centuries (it won’t), I would expect most roadkill species to adapt.
And yet I’ve seen many times a bird flying well above human heads suddenly swoop down to car-radiator or car-window :eek: height while flying across a street. 
When I lived in Bozeman, there was a de-facto dog park in the middle of town, with houses across the street from it on all sides. I once saw someone open the front door to let the dog out, and the dog then walked over to the street, looked both ways, and then crossed over to the park side to play with the other dogs.
I would say that in heavily forested areas roads are a net benefit to deer populations. They provide a lot of edge and edge is great for creating habitat and diversity. The volume of food in an acre of right-a-way is probably a hundred times more than a mature forest. They have access to salt if they salt the roads in the winter. They have the cover and thermal insulation of the forest but when snowpacks are high they can walk easier on the roadside and sun themselves on the blacktop. Many people don’t realize how tough winters can be on animals. It can come down to a few percent difference in fall body fat or an animal having to burn a few hundred more calories a week.
So,
The net fitness equation may be greater crossing the road than avoiding crossing the road (and likely is as simply limiting their ranges by that magnitude would be of major impact), a few species have the intellectual capacity to develop cognitive and possibly even culturally transmittable avoidance tactics but many do not and their extant tactics (run fast, have shell, taste bad, have a very noxious emission, freeze …) do not adapt well to this threat, possibly adapting those tactics to this threat may have unintended negative fitness impacts in other ways, and maybe it is having a selection impact at a rate that is not easily observable at the time scales we are observing.
On a related matter: the typical response by (wildlife) officials to a serious or fatal attack by a wild animal on a human is to trap/hunt down and kill the offending animal. Case in point: in the news today from Australia there was an unfortunate case of a demented elderly woman who wandered away from a care facility; her remains were discovered on a riverbank under circumstances pointing to a crocodile attack. Authorities are busy setting traps to catch and kill the croc on the grounds that it was was “bold” and aggressive (and thus allegedly more of a threat than its calm, peaceful brethren).
You might think that culling of this sort would select against nasty critters that attack humans, but it doesn’t seem to have worked out that way. Crocs, grizzlies, giant snakes etc. keep right on killing and consuming people when opportunities present themselves. 
Actually, most big predators do avoid humans most of the time. I would guess that’s because we’ve spent thousands of years tracking down and killing the ones that don’t avoid us.
Probably works better with smart predators, like lions and bears, than with relatively dumb ones, like crocodiles.
Or it could be a social behavior. The humans understand the traffic pattern and cross the street when it’s safe to do so. The cats and dogs don’t understand the traffic but they move when the other animals (the humans) move. They don’t have an instinct to avoid traffic but they do have an instinct that there’s safety in staying with the pack.
I feel like the local deer are better at avoiding cars than just a decade or two ago. Just last night, I saw a deer on the side of the road. It looked at my car and dashed into the woods. They used to freeze or jump in front of cars.
That’s probably learned behavior, not evolved behavior. Or it may be a little of each.
You can also “learn” things through evolution. Imagine a population of (say) deer, which all have random phobias. Some of them are afraid of the full moon. Some of them are afraid of spiders. Some are afraid of green eleven-toed hedgehogs (and if it ever saw one, man, would it freak out). And some are afraid of humans. All of these fears are genetic to at least some extent, but have no particular reason for them. But then, suppose that humans move into the deer’s habitat. The ones who are afraid of us would stay well away, but the others will come right up to us idyllically with Rossini music playing in the background. And then we kill the ones that come up to us and eat them. The ones who feared us and stayed away will have much better survival rates, and so the next generation will have a much higher proportion of human-phobic deer, even without any individual deer ever actually learning anything.
From what I saw on a 3-week roadtrip in Australia, kangaroos have a lot of learning/evolving to do, roadkill-wise.
I’m not sure what distinction you are drawing here? This is a description of the way all instinctive behavior evolves. Genetic variation, arising through random mutation, specifies a range of random behavioral phenotypes; if one of those genetically-determined behaviors turns out (by chance) to have a fitness advantage, natural selection favors it, and the genotype (and associated behavior) proliferates in the population.
The distinction I’m drawing is between evolution of an instinct in a population, and learning (and possibly subsequent teaching) by individuals within that population.
Over a million human beings die in traffic accidents every year: (http://asirt.org/initiatives/informing-road-users/road-safety-facts/road-crash-statistics)
Natural selection doesn’t appear to be improving that statistic!
I’m not the person you asked, and I’ve never counted bodies or mass in the road, but I’ve got 35 confirmed (as in I checked they were dead) road-kills (yes, I count 'em), and I’ve got 2 cats, 2 seagulls, 1 squirrel, 1 raccoon, and the rest were all small birds who’s species I never bothered to establish. I’ve successfully braked/otherwise evaded many more, including at least one deer. Including the successful misses, body numbers are going to massively favor the small ones, which will be too small to notice or be eaten before they’re noticed and counted, but body masses actually seen dead are going to strongly favor the big ones not eaten or otherwise cleaned up quickly. It’s not something that’s going to be establish-able without a major scientifically designed daily ‘road-kill census’ project on large stretches of carefully selected representative roadways.
I can think of a couple of instances where evolution would work in favour of increased roadkills.
In Australia, many a visitor’s first contact with a range land kangaroo is seeing one hop onto the road 20-50m away and then be transfixed by the oncoming headlights @ 120kph.
This is a misfortune for any kangaroos struck by the vehicles and consequently a continual income stream for the smash repairers.
The reason for kangaroos being so prevalent on the roadsides is the structure of the road itself; the verges catching and holding any run off from whatever rain falls. As any drover will tell you in summer the forage in “the long paddock” is superior in both quantity & quality to what can be found a safe distance away from the roads.
So animals who learn to forage along roadsides may have greater survival than the more timid individuals, at the cost of a proportional loss as roadkill.
In a similar vein, the harvesting and transport of summer cereals i.e. wheat, oats, barley, sorgum etc from farm to local grain elevator/silo inevitably sees spillages along the roads which feed flocks of thousands of galahs, cockatoos, corellas etc at the cost of a few percent likely survive the summer better than flocks who watch the alternating feast and carnage from a safe distance.
There is another side to this. In the UK, the wide verges, largely uninterrupted by side roads or other obstructions, are a valuable corridor for wildlife. I see far more roadkill on country lanes than on other roads where speeds are much higher. This may be due to many factors, not least that I will be concentrating more on driving and less on identifying the ‘litter’ on the verge. The most common roadkill I see are badgers but that may be because they are quite large compared to the millions of smaller species that probably die on roads every year.
Animals do learn, and we know that some animals are born with innate behaviour and usually learn from their parent(s). Sea turtle hatchlings instinctively dig their way out of the buried hatchery at night, when they are safest. Honeybees learn to associate given colours with the food they’re seeking - it is not innate.