Well - in theory - this could be caused by flaky bus buffers (74245) and/or their _OE signal. Provided you've double checked the PSU(!) and ruled out heat issues, you could use a logic analyser to check the _OE pins (19) on the 74FCT2245 chips for any spikes when the board is deactivated.
Do you have any chance to test the board in another 2000?
Are the error bits always the same or is there any pattern in the errors?
PS: make that a logic tester, no analyser ;-)
