AMD ships low-cost six-core Phenom IIs

Parallel software

Make: (builds software projects and can split the task over a large number of cores). Compiling software is cache intensive. At some point, adding more cores will not make the build faster but adding more computers can. The first time you build a large project, you might have to wait a couple of minutes, but on subsequent builds, Make only rebuilds the parts of the project you have modified.

Apache: Can put separate http connections on different threads - if you have enough visitors. Again it is possible to distribute a web server over multiple computers, and doing that can be more efficient than having more cores.

My home machines have been fast enough for a long time, and are now optimised for silence. The next useful step will be to cut the number of cores to the point where the CPU is cold enough to put DRAM dies in the same package as the CPU. (ARM did this years ago). Much of the energy wasted by AMD/Intel/Via machines is used to send high speed memory signals through a chip socket, across the motherboard, through some DIMM sockets and along all the DIMMs.

I am surprised combined memory and processor chips are not already used on graphics cards. ATI-AMD and nVidia make separate expensive and cheap chips for noisy and silent graphics cards. The expensive chips are too hot for DRAM dies, but the cheap ones should be fine. A few cheap combined graphics/DRAM chips would be just as fast as one expensive chip with separate DRAM. Smaller chips have a higher yield than large chips, and cutting the number of different chips cuts down costs too. The graphics market is still competitive enough to be driven by performance per dollar (certainly compared to the Intel market).