Even if KnC/Bitfury/etc. were >50% and dishonest, the socioeconomic majority can flee their corrupted PoW by forking to something besides SHA256.
Not clear that isn't just a treadmill though. If SHA256 became corrupted than given the same structure something else might very well become corrupted too.
Various arguments could be made about difference in absolute time, relative ASIC-resistance of the function, etc. but I'm not sure how compelling they are.