Why and how would they do that though? They can't just go and change it like that. If he solo-mined, he would just have to have a version of the miner that didn't have that requirement.
This is possible. You can see it in action with the new merged namecoin/bitcoin implementation. As of block 24000 it'll be a slightly different algorithm for namecoin mining. The client will treat blocks < 24000 ('old' namecoins) differently than 24000. And an old version of a client would simply be incompatible with the network -- you couldn't perform any transactions until you upgraded. After being sent the 'old' coins would become 'new' coins, in a new block backed by the new algorithm.
So changing an algorithm to torpedo ASIC miners is very much possible and could happen if a single ASIC miner farm threatens the whole ecosystem.
That's why I think ArtForz will release his chips to the public, or at least promise to do so after some period. Not doing so may force the bitcoin devs (who are mostly early adopter types with substantial BTC holdings) to do a bit of fancy footwork to protect the ecosystem.
The memory intensive proof of work algorithm Joel K mentioned could be interesting. It'd swing us back to CPU mining and level the playing field somewhat. But the hard part would be balancing it vs. the diffuculty of the time. Crashing the network hash speed by a factor of 10000-100000 would not end well.
Oh yeah, like everyone I'm thinking now is a good time to cash out. I'll hit the local craigslist and see if I can sell these XFX 5830s while they still have a ~$100 value and grab a 460SE or something until the 7 series pop. If nobody bites I'll raid microcenter and offload them in a cheap crossfire gaming system with the spare win7 pro license I have kicking around. Under $13/btc is getting uncomfortably close to the "so not worth doing even at 6c/kwhrh" level.