After fixing a bug that let the new server grow to 2.5GB RAM, it is now stable at 500MB RAM. I also discovered I can't apply patches live and save the image, because that means its saved with the RFB server running.
This large bulk load uncovered a bug when using Magma in direct-mode since Eliot fixed the process-scheduling to not cycle through equal-level processes by default -- Magma's flush-to-disk process was at equal priority and so it wasn't getting flushed! Spur really needs a way to cap its memory allocation like the old VM's had, so something like that would not cause problems for other processes on the server..
Disk space is so tight, I'll have to log in to delete unneeded commit log files to avoid us running out. Hopefully Rackspace can afford us a bigger disk.