Here's the problem for me. These mining operations can be exploited to an extreme. Any scientist or professor with knowledge in protein representations will have an unfair advantage over normal miners. This gap will be even larger than the hashpower gaps we see between ASICs and "normal" miners in bitcoin, litecoin, etc because scientists will have access to not only hardware but also extremely specific domain knowledge.

The problem with solving "real" problems instead of synthetic ones is that they aren't uniformly difficult, so clever people will game the system.