That's a good point. Most of the programing here is done in C right?

Any of it done in C++? NVIDIA has been working on C++ compilers for CUDA so you can program it in C++ and run it on a CUDA enabled device.

Just thought it was worth mentioning!

Not at WCG, maybe at other GPU utilizing projects.

@Otis11 - OpenCL as WCG has no GPU projects (yet).

Now may not be a good time to introduce CUDA.

The WCG is busy. The scientists have just finished 3 projects and are readying one or two more.

The technology is also developing and their might need to be more commonality between ATI and NVIDIA before it will be worth pushing out tasks from here.
There are many practical issue to consider, such as variants of card capability to use, which could change soon. Most GPU tasks from other projects have stopped using Compute Capable 1.0 cards and now only use 1.1 or 1.3. The Milkyway project has opted to only use CC 1.3 cards. That is a GTX260 or better! They are not cheap, so the number of users is quite small relative to the number of CPUs capable of running Boinc and WCG tasks.
Dont think these top card numbers will grow either, they will not! NVidia have stopped manufacturing all but the GTX295, in preparation for their new range (perhaps out before the turn of the year). Who knows what they will bring, and you would be somewhat foolish to produce tasks designed to run on CC1.1 on a 2 or 3year project! By the end there might not be enough people using the cards.

Although NVidia are still producing lower powered cards, all CC 1.1, their performance is no-where near that of a top end card, so there really is no point in writing code to run tasks on a graphics cards that are not any faster than a CPU.
Take the Nvidia ION or the Geforce 8600M for example., These have a Boinc rating of 6 and 5 GFlops respectively (not the same as other rating systems). These would not be faster than an good quad core CPU. In fact a GTX260 would be about 20X as fast as either.
Then there are the mid ranged cards, the 8800s and up, or the more eco cards. They all present issues to the programmer.
Card use in itself might make a complete mockery of the badge and points systems. The badges would lose any meaning, so what would be their point?

Dont get me wrong, I am in favour of GPU computing and support it, my one GTX260 does at least 3 times more work than 3 quad core CPU systems, and I have other cards. It is however difficult to do, frustrating to implement and might not be worth it, just yet! The WCG runs lots of projects and most are not adaptable towards GPU computing, at the minute. They need to continue to support the existing projects before going down a different road. When GPU computing does develop further, and it will, it might be more usable for projects at the WCG. Let the techies and scientist that actually run the WCG decide what is best for the WCG, they know about it too! For now turn your cards towards Folding@home or the GPUGRID. They do good work too, and are GPU specialists.

Actually, the fact WCG *only* supports cpu's is the reason I run it. Notwithstanding the benefits of cuda, etc the badges/points are more meaningful from a comparative standpoint when universally cpu driven, at least within project. The point system for Boinc as a whole is all but broken, no sense breaking it even more.