Re: [8500 GH/s] Slush's Pool (mining.bitcoin.cz); TX FEES + UserDiff; ASIC tested
Well, the negatives were amusing. I suspect that what happened is the numbers got bigger than could be represented with the number type in the code. when that happens the sign bit gets changed and the number wraps around becoming negative. basically the biggest number + 1 = the most negative number. 
I already PM'd Slush about my datatype overflow suggestion. He hasn't been only since. But it's nice to see others have the same idea as to what may be biting the pool.
Hopefully it can be identified soon, as that could lead to a real solution and less work for the operator repeatedly having to recalculate certain blocks.