so destroying 512 epair(4)s is... surprisingly slow, if you do it inside a vnet. there seems to be some sort of bottleneck (presumably around locking/NET_EPOCH) that means the kernel can only destroy 3-4 of them per second.
probably not something you would ever do in reality, but it makes the if_bridge(4) tests fail if you run them in a jail since the cleanup times out waiting for that.
i am already extremely suspicious of the correctness of the net/interface locking. hmm.