After a brief discussion with Rusty about hackbench he explained that it 
was more a pass/fail benchmark than anything else - failing by running 
forever. Well recently I noticed all kernels were doing it so did some 
regression testing to see when it started. 2.6.7 runs fine, but 
2.6.8-rc1 onwards get hard rebooted after 12 hours.

I've been using the 8x boxen at osdl with
TEST: hackbench on stp8-001 running PLM ID 3049 [ linux-2.6.7 ]
LILO: append = 'profile=2'
Script Params: 25 200

I'll investigate further to see if I can track down the patch that 
started this but it may take some time. Suffice to say that the 
2.6.8-rc* versions may not be safe in certain server settings until that 
time.

Cheers,
Con