pion: I examined a number of the dump files you sent me and most of them are hung in closesocket. This is familiar territory for me. I eventually figured out that what was happening is we passed a socket handle onto a child process (which is easy to do unfortunately) and then tried to force the socket closed. This can result in the server hanging until the child exits or perhaps blocks in an alertable state, but scripts don't expect to need to do this and so they often never do. I discovered this by accident back in late v6 and fixed the server to protect socket creation. It wasn't until v7.0 that I realized I also had to prevent the TCL library from creating sockets while ioFTPD was creating processes. So as I mentioned above I had a thought that libmysql probably has the same problem which is why your configuration is different than others and they aren't seeing any problems...
Anyway, I setup nxMyDB and performed some tests against the modified library that I thought should have fixed this. Evidently I didn't do something right because I can confirm that it looks like a socket handle is being passed onto child processes by nxMyDB but which really means libmysql. Debugging through several layers of dynamically loaded libraries really sucks, but I'll figure it out for you.
I can't prove this is the problem, but given past history I'm pretty sure this is it. So there is hope yet
