The QMD problems require 512MB to 1 GB of memeory to run, so you will need plenty of memory.
They apparently give you extra credit in the points department since these problems use more system resources.
Due to Stanford using an Intel compiler, these problems will run faster (2x reportedly) on Intel hardware with SSE2. Some AMD cpus have SSE2, but the Intel compiler doesn't take advantage of it. At first, I read that they were not giving these problems out to AMD computers, but it looks like they are as I found reports of frame times for an Athlon XP. More here QMD FAQ
So if you have an Intel CPU and lots of memory you should be able to rack up the points. You will need to configure the client to accept big WU's and use advanced methods.
I haven't tried these since all my Folder's have AMD CPU's, but I would be interested to hear if anyone else runs a QMD.
Interesting, since GCC 4.x generally produces code that is competitive with the Intel compiler speed-wise, without the Intel bias. The Pathscale AMD64 compilers (popular with high-performance apps) can suck an amazing amount of floating point performance out of AMD hardware, but you have to pay for the privilege. Most of these problems appear to be better suited to AMD than Intel, based on a loose survey of typical parameterization. Same ISA, significantly different design choices.