Dmitriy,
ICC 10.nn.dont.remember using /O2 or /O3. I haven't retried the problemcode with the current version. The problem was you debug with no optimizations. When that works you start increasing the optimization levels. When you optimize to the max and code works for a long time under rigorous testing you think everything is fine until some time later when the race condition is encountered. Then after a day or a few days of back tracking you come upon a new circumstance where the optimizer failed to produce the correct code. After that hits you a few times you tend to be a bit on the cautious side.
I will be using manual hoop jumping since I cannot rely on the problem creeping back as optimizations get more aggressive. (IOP is a bit too aggressive)
Jim
