This might be useful in terms of test tools. Cluster people care a lot about getting their systems reliable (if you had 512 PC's you would too) and there is some good info in that posting about stress tools including va-ctcs/cerberus and the like.
http://www.beowulf.org/archive/2006-October/016669.html
For I/O stressing take a look at
http://samba.org/ftp/tridge/dbench/
and the related tbench tool for network stress.