I'd disagree with slpLoad - yes, individual machines can work by themself, but through XGrid etc, it's a potent tool to create dynamic clusters of computational power to do work on, if you have the right kit.
Possible uses: http://www.nvidia.com/object/cuda_home.html#
I know Server can pool resources, and I think you can set up an XGrid by commandline/tool for non-server versions for using other Macs' resources.