Well it’s a good thing I never studied law…
In principle a microwave repeater could have latency little more than the length of wire inside them. However you would be wanting to perform some massaging on the signal, usually moving the carrier frequency so that you are not transmitting on the same frequency you are receiving on. That will add a few cycles to the transit time, but at these frequencies, still very minor delay.
Where you will pay is when you have a large LEO swarm, and you need to switch or route between satellites, not just repeat. This is where packet switching starts to win, as you can set up virtual channels, and the packet carries the switching information in its header, allowing near wire speed switching of the data. The incoming packet only needs to have that part of the header defining the outbound channel read before it can be on its way out of the switch. Moreover, internal operation of the switch can be operating on a per bit level of the channel identifier. The downside is that communication requires setting up the entire end to end channel before communication happens, as the sender needs to know the switching. Latency can be as low as the number of bits needed to represent the identity of the next hop.
Routing tends to need the end point address in full, and then to lookup the routing information to determine where to send the packet next. It can still do this once it has the header, but you pay a few more bits of latency, and the cost of routing table lookup may slow things.
Either way, throwing enough smarts at the problem, you can get transit time through a data satellite
down pretty low. In general you would want the sat to be a simple as possible. So packet switching makes sense. Lots of store and forward is not the way to go. But the dynamic nature of a LEO swarm will make a lot of this a much harder problem. The set of ground stations and other sats in communication range will be constantly changing, and will the routes taken. Static virtual circuits won’t work, and routing tables will be very dynamic. Much fun to be had to say the least.
I can think of a number of interesting ways of going about this, I suspect some of the finest minds in the business have been all over his for quite some time. Probably some highly proprietary solutions.
Better to use tachyons which would (theoretically) reduce ‘ping’ times to negative numbers.
As you say, for some definitions of “care about most” of course.
Air raids and stock markets also come to mind.