I work for a film VFX company and we utilise several hundred Linux machines to render all of our work. This can involve using multiple machines to work on one single frame, or to render a batch of frames. We can't utilise all the memory all at once as you suggest, but by splitting the job into smaller chunks then we can get the work done much quicker :)
There are (at least) 3 on-line distributed computing sites that work by using your spare cycles:
The most famous is seti@home which is looking for inteligent signals from space http://setiathome.ssl.berkeley.edu