The same thing is happening to my Mac pro tower (late 2009 running Snow Leopard).
I have done some reading of people with the same problem. Some say it has to do with the way the snow leopard allocates the memmory, and recommend doing a permissions repair, showing it reduces the active memory usage in Activity Monitor.
Link Davis recommended using the terminal to inset a command to find out more:
syslog -k Sender kernel -k Message CReq 'GPU D|I/O|find tok|n Cause: -|timed? ?o' | tail | open -ef
and to see if a message appears in a separate window after it has run.
In my case it did:
Fri May 16 21:12:25 Joes-Pro kernel <Debug>: 00000000 ffffffff Intel82574L::timeoutHandler - link is down
Fri May 16 21:12:36 Joes-Pro kernel <Debug>: Previous Shutdown Cause: -128
he said a negative number could indicate some kind of hardware failure (very vague)
I have tried all these things including starting up from a different internal drive, starting in safe mode, holding down the option key and starting from another drive, and the restarting problem has occurred during it all. Finally after pushing in my power button and forcing a shutdown and then zapping the PRamwhen it started back up stabilzed it temporarily.
I've had the machine at an Apple certified repair and diagnostic place and they couldn't find anything.
They suggested replacing the drive(s), but I have done extensive low-level scans and no bad boot blocks have been found. The hard drives pass the tests in Disk Utility