I agree it’s seldom a big difference when looking at all possibilities 3+ months away from the election that you’d pick up 2% in one state. But few other people she’s considered even do a little to get a particular state (Sherrod Brown of OH was one, but Kasich would name a Republican to replace him, McAuliffe a Democrat to replace Kaine, assuming the ticket wins). And let’s say VA is Clinton +2 points, then a 4-ish point national drop by Clinton is needed for Kaine to just make up the difference, but he starts delivering the state at the margin with a 2 point drop, she just barely loses without him, wins by 2 with him, hypothetically. We don’t even have high confidence polls reflect reality to within 2.
I also agree VA is not as likely to be the last piece of the puzzle for either side as some others, but it would be key in some plausible scenario’s. If Clinton only loses the Obama states of FL and OH, NC was a Romney state, she still wins. She loses if Trump adds PA, not won by GOP since Bush I, but his most likely critical path to victory IMO, assuming again he holds all Romney states. If he can’t do that, then he needs to add NH, IA, and VA as most likely candidates (CO couldn’t be excluded). So boosting in VA is worth a bit to block the second most likely critical path for Trump.
Partly relevant point on 538’s interesting graphs and %'s, among those I find less plausible are very small % likelihood EC result is exactly the same as 2012, seems to me an artifact of volatility/correlation assumption using past results in different situations. IIRC the latest output put it like 1% chance, I would think it much greater than that, at least 10%. Likewise in general I’m skeptical of other 538 results relying on volatility/correlation assumptions. And there’s no way to gather enough real world results with a known stationary distribution of outcomes to ever conclusively say eg. the chance of X particular was really Y%, he was right!