Looks cute, big part of that is the GPU though, not the CPU. And IMHO, even a quad-core processor won't be enough for doing nice physics (where objects
really influence each other), you need a PhysX card for that...
I wonder how high their CPU utilization is, as it's a pretty complex task to do parallel programming - especially for games, which have a lot of interdependent systems. Ie., you can't fully do AI before you've done physics, and physics can't be fully parallelized either (although if you're clever, it can be done as an iterative process where you parallelize subtasks and re-do when there's object interactions), but imho physics should still be moved to a discrete component.
You might want to check these links:
http://www.anandtech.../showdoc.aspx?i=2868http://arstechnica.c...valve-multicore.ars/