I decided on using 10 of the beasties, run from an old AT power supply and connected by an old 10MBit hub. First came flashing all the units. I combinded Linuxbios with Etherboot, which resulted in a device which could, from the BIOS, load a Linux-kernel from a server and boot it. The linux-kernel in its turn could then mount its root from NFS and start running the userspace programs the distro provided. That way, all I had to do was reflash the BIOS-chips of the WT300 machines, without needing to solder an IDE-interface into each and every one of them. The distro I decided on running on the devices is a lightweight one made by uclibc's buildroot, and included gcc and distcc, a program to split a compile amongst multiple machines.
On the hardware side, I needed a power supply and a way to distribute it.
The power supply would be an old AT-PSU which could deliver the 10A
on the 12V-line the terminals would need. To make it a bit easier on the
eye, I modded it so only the 12V-line and the switch-cable would come out:
The speaker-cable I had reserved for delivering the power to the WT300s didn't fit through the hole of the cinch connectors I bought, so I had to connect them another way:
And finally, lots of UTP to connect it to the 10MBit hub:
Turning the cluster on revealed it worked, but booting took some time: because the machines were all connected to a hub and tried to boot all at the same time, it was collision galore. Luckily a friend still had a semi-broken (it tends to overheat) 100MBit switch lying around, and with that, the cluster booted way faster: