I haven't seen this problem myself. So there's a catch here. Could you submit a bug report after it fails? The report might yield some interesting clues.
A few more questions to help narrow this:
- Does the operation actually set a visible breakpoint?
- Can you describe the test run itself in more detail? Does the test show as running in the Processing Queue? Does it hang?
- If you try to do this 5 times, how many times out of the 5 does it fail?