I run exclusively Projects that are part of the BOINC (Berkeley Open Infrastructure for Network Computing - for all non Distributed Science folks) framework. Folding@Home once considered writing a BOINC compatible client but eventually rejected it.

It did itch in my fingers to try the GPU speed of my X1900XT, but their GPU Clients had Beta and expiration dates written all over them... plus they each required specific (usually older) Catalyst Driver versions which weren't suitable for everyday's Gaming needs.

The nail in the coffin were the power requirements and subsequent constantly very high GPU temperatures that were to be expected. From what I heard, running the GPU Client was worse than running a 24/7 3D Benchmark - stressing the GPU cooling to its maximum. Additionally, it still caused very high CPU overhead, thus making the CPU resources almost unavailable to e.g. other data crunching.

--------------
GPU crunching is coming along slowly but is making progress.
Eventually (within ~1 year from now is my guess) we'll see some Distributed Science Project making regular optional use of modern 3D Hardware.

Just a few weeks ago, someone compiled and ported SETI onto the NVidia CUDA platform and keeps optimizing it.

Overall, I find the idea to crunch data at highspeed off a GPU pretty sexy, but it has to run with any modern Driver (not just a fixed version), plus it needs to work in ways that do not interfere with normal Operations (the way it is now, the GPU Clients can make the System unresponsive and they won't shutdown/pause when 3D Applications are launched - causing erroneous System behaviour).