My question is: How bad is this? Is it terminaly bad or simply not so great?

Asymmetric routing by itself has virtually no impact on performance. It does present a measurement headache, though. This is discussed in a paper of mine available from ftp://ftp.ee.lbl.gov/papers/routing.SIGCOMM.ps.Z a revised version of which will be appearing in IEEE/ACM TON later this year.

Out of order TCP packets are bad, but can simply be reassembled.

Asymmetry doesn't lead to out-of-order delivery, so this isn't a problem. Routes that *change* often, on the other hand, can indeed lead to performance problems. Vern