I wouldn't go so far as to call reCaptcha a gimmick. It is an excellent example of how a slight modification/re-mapping of the original problem-space can solve more socially/commercially useful problems.

PoW has proven itself in the field for years, while PoS is still untested at scale. Being able to piggyback scientific/medical/environmental computation over a PoW scheme would provide significant value as opposed to a fairly useless compute problem such as the double-SHA256.

The key here is to be able to re-interpret/re-state a problem in terms of the PoW algorithm with the following characteristics: - Expensive to calculate, cheap to validate - Dynamically-adjustable difficulty

An ideal implementation of this would be similar to an OS running an idle process (double SHA256), and switching to running user-processes (medical/scientific computation) when needed/instructed.