Heres my take on it!
While generally it is true that a brand new chip sporting a reasonable tempurature and reasonable voltage level will over time become more OCable as it becomes burned in, eventually it will probably get worse someday if it doesn't just suddenly poof, there are always exceptions to this and its possible some of the weak transistors in that chip have become even weaker. This is probably unlikely as the majority of results are the former as opposed to the latter when well planned overclocking is involved.
Maybe its possible that the cpu is getting too hot, this could cause the latter. On the other hand, its possible it isn't even your CPU... maybe your chipset is getting too hot, there are many different possibilities really. If there are rounding errors in the current conditions, then its best to stay away from using it in a way that results in rounding errors. You changed the board and everything so yes it looks like the cpu, I would still want to be redundant and attempt to test out the new board too, maybe you can run a program that works the chipset and fsb and ram hard while going easy on the cpu. what is your agp/pci frequency at? Locked? Do you notice a different in temps between when it could do 3.3 and when it failed the test at 3.3?
Well at any rate I hope you find out.