I had an issue with our medical software this last week that I could not reproduce my knew it was happening in certain conditions. I just kept on it until I finally made it fail...then I figured out how to make it fail consistently. Once I had this issue in a repeatable state it was a relatively easy fix and I knew it was a reliable fix.
I know this is all stuff you already know...but I think the first order of business is to reproduce the problem consistently.