Don't forget to factor in the cost of electricity. Make sure you look at the total cost of crunching and how efficient your setup may be.
Cheers

Although many BOINC projects are behind the curve, GPU computing will get you much more computational horsepower for your buck and energy consumption. HFCC doesn't have a GPU client; HCC does, and I'm sure that there will be more sub-projects that will want to take advantage of GPU computing soon.

Obviously, a more powerful CPU is going to get you more points if that's your goal. Number of cores is also going to be a factor with CPU computing, but make sure you have enough memory per core (at very least 1 GB per core, preferably 2 or more) to keep the single-threaded applications happy. Don't know if WCG is looking into multithreaded apps or not.

While there are those who spend a lot of money to get an edge in grid computing, my personal theory is that people shouldn't be bringing a bunch of hosts online just for that purpose, as it's rather wasteful of resources and energy. After all, the idea behind grid computing is to decentralize the work, so one group doesn't have to stock and maintain a large farm of computers.

We'll ignore the fact that I currently have six hosts (12 CPU cores, 3 GPUs) in the mix running 24/7 for badges on short-term projects--that's not something I normally do. In terms of badge hunting, actual performance isn't important as it would be with ponts--just the number of cores you have running, since badges are time-based. I have an old G3 300 MHz iBook running right now, just for badge purposes. One could say that it's even helping the badge situation, since it's not thrashing through work units as soon as they are sent.

Thanks for all the information :)