A few clarifications
I just wanted to reply and mention a few things about the plug cluster (as a member of the team that got it working). We had a total of 17 sheevaplugs (16 compute nodes and 1 storage node with a USB attached storage) all being managed by an EEE PC 901. All of the nodes in the cluster were running in diskless mode (their root filesystems being served via NFS from the management node). The display on the EEE was showing the output of ganglia monitoring the system as we were running tests across the system. While we didn't have cpu-intensive jobs (or applications) we were running the Intel MPI Benchmark suite, which showed quite well that the systems were able to handle an MPI load effectively and correctly (the suite was run continuously over the course of the week and reported no errors during that time).


ARM+GPU = dream super?
Nvidia's Maximus: Hard-core 3D graphics on speed