I have a concrete suggestion. There are several combinatorics-related sequences I can think of that are well suited for highly parallelized distributed computing, but for which only a few terms have been explicitly calculated due to exponential computational complexity. Here are two examples: OEIS A088672 and OEIS A028420. I have already written fully parallelized and optimized code for both of these; other sequences which I have in mind would be similarly easy to code.

Actually, this summer I had intended to create my own BOINC project along these lines. One of my friends was supposed to take care of the server side of things, but he fell through on his commitment. So the idea hasn't gone anywhere yet.

Here's an idea though. If someone wants to work with me to get a project up and running (I manage code for workunits, other people manage servers), then that project could be used as a foundation for more ambitious objectives, like the 'MicroGrid' concept I've heard mentioned before. The key point would be that the combinatorics problems I proposed would provide a steady and inexhaustible supply of workunits.