Comment 10 for bug 1038167

I was able to review this with Alex today.

I think Chris Rossi's analysis isn't 100% correct: the ZEO cache layout assumes that other agents can operate on the directory at the same time and has appropriate locking and waiting behaviour. I can see that relstorage also implements the same protocol.

From my perspective we're tracking down a bug here, so I think we should

* identify the cause of this traceback (I don't understand which part of the application causes the traceback)
* fix the bug
* stay with a single directory
* add external tmpwatch