Marty Shapiro writes:

Care to put some money on that conclusion?

I have no money to spend on gambling, but I'm confident that the
conclusion is correct.

Many, many years ago we
encountered a similar problem. Every symptom said it was software,
especially when one operating system would not encounter the problem yet
another one did. After several weeks we managed to isolate the exact
channel program sequence which caused the error. The hardware developers
confirmed that the channel program sequence was indeed valid. Further
investigation showed it to be an error in a circuit card supplied with a
field upgrade kit. Since each operating system used a slightly different
channel program to accomplish the same thing, the error only manifested
itself on one system.

Those are the stories one remembers _because_ they are such remarkable
exceptions to the rule. I think the explanation in this case is much
more mundane: bad code.

--
Transpose mxsmanic and gmail to reach me by e-mail.