If the PoW function is changed, it ought to change slowly so as not to drop a brick wall in front of the miners speeding toward the ever-receding goal of protecting the blockchain. Who's going to get on that path if the bitcoin community does that?
But it can be done slowly. If most of the entries is the list of possible PoW functions are double-SHA256, then the few that aren't will offer the healthy goal sought by those who like the idea of changing it. The healthy goal is for general computing machines to help protect the blockchain in an incentivized way. There's a sick goal too, which is to destroy large investments in mining. I hope no one has that goal.
At
http://bitcoin.stackexchange.com/questions/35679/is-it-possible-to-make-pow-asic-resistant-through-dynamically-generated-hash-cha/40475#40475 I proposed that ongoing competitions for the creation of new hash algorithms could feed an ASIC-resistant PoW, defined using the as-yet-unknowable winners of such competitions. It is possible to make an ASIC resistant algorithm, but it isn't a programmable algorithm - it's one that requires human intervention. The hash of the next block is a good example - there's no programmable algorithm that can find it because too much human intervention is required, but it's an algorithm well-enough defined for us to build a billion dollar system on top of it.
That being said, I've started looking at two different kinds of decentralization. The literal actually-in-different-places kind is categorically different than the much more important, virtual impervious-to-coercion kind. The behavior of the "centralized" oil cartel is a good example. The participants cheat. This is a fundamental principle in the debate between free-marketeers and authoritarians regarding the emergence of monopoly. Without coercion, monopolies fall apart. There's nothing coercive about our use of the double-SHA256, so in my mind, the centralization it has so far produced is not dangerous. It's scary, sure, but until coercion is used to prevent me and my friends from buying our own ASICs, it remains impervious to coercion.
Sorry for the long email that didn't make any apparent progress. The thinking is what matters to me, and seeing two kinds of decentralization and recognizing that a change in PoW can be slow enough to avoid hurting existing miners are items I haven't seen anyone else recognize, so I had to bring them up.