3DMark2001 Failure - High Polygon Count

It seems I've fallen victim to my own P4C overclocking theory posted in this thread:

http://forum.oc-forums.com/vb/showthread.php?s=&threadid=209759

I can run up to 270 FSB making the adjustments recommended in the thread. However, 3dMark2001SE consistenly fails only on the High Polygon Count test. Specifically on the 8 light test. The other tests and Prime95 run fine.

Failure occurs around the 15th or so loop. I run a continuous loop with just this test selected. When the test fails, the system reboots by itself.

At 260 FSB, I ran 50 loops without any failure. 265 FSB failed.

This is a bit strange. Anyone have an idea why the polygon test seems to fail after running for a few minutes? Does anyone want to test their own system?

Yeah I use 3dmark2001se to check my system as well. My system crashes at that point too when I have my fsb at 276. But at 275 I can run that all day long and nothin would happen. Prime95 stable as will.