I'm all for "ScienceCoin" concepts, as long as there's a good POW function. Centralization is good in the case that a party can enforce the use of a certain algorithm, and implement checksums to make sure that the appropriate work has been performed, but is against the grain of most cryptocurrencies current out there.
It's true that some centralization will be necessary for a project like F@H because they have to provide the network with new data to work on once in a while. The proof of work algorithm on the other hand can be centralized or not depending if it's possible to implement something that works for all future proteins. Decentralized is of course better but, I would still "mine" for a project where the proof of work algorithm is manipulated by a central authority if I have trust and respect in the said authority. I know that not everyone trust universities but I personally trust them more than banks! The most important part I think is not that it's centralized or not but that participant can't cheat and that the project offers total transparency.
I think that if we wan to get a lot of miners into this, we need to find the right scientific project. We need to find a project that is only based on pure maths and where the proof of work would be undeniable. The only project that comes to mind is the PrimeGrid project where their goal is simply to find huge prime numbers, this is similar to hashing, no? There are probably other good projects to look into. The BOINC project page is a good place to look to get some ideas going:
http://boinc.berkeley.edu/wiki/Project_list