As far as I know, all competitions on Kaggle are on basis "who is the best". So there is no limit on skills; and actually if for someone a problem is trivial, then one can safely claim the prize (e.g. Heritage Health Prize).
From my point of view it allows both to enter interesting data, compare how one's techniques compering with others and learn other approaches (usually winners disseminate their solutions).
When it comes to the only thing in the academic world which is considered to be a serious stuff, yes, some works end up as publications, see Academic Papers - Kaggle.
Also, some competitions are research-centric (however, with no cash prizes), e.g. Eye Movements Verification and Identification Competition.
This in an official competition for BTAS 2012 (The Fifth IEEE International Conference on Biometrics: Theory, Applications and Systems, September 23-27, Washington DC, USA) and all results will be published during that conference (and of course on this web page as well).
However, there are various competitions, so I guess it's hard to make a general statement about "research-worthiness". And for all what matters the most is the result, not its 'purity'.
Out of my personal stuff:
<spam>
I made an entry (a graph map of tags) for Kaggle StackExchange visualization competition.
</spam>
Perhaps I'm starting in DarkWorlds, but more for fun and to practice machine learning techniques, as my field is neither machine learning, nor astrophysics.