The 'Server did not stop' error was lying in wait for the NoneType bug to be fixed :)  So, if you didn't the NoneType, my test case isn't reproducing the problem for you, which is odd.  Using call_soon_threadsafe from an addCleanup in the _server method makes it work for me.  (Seems to work even without the threadsafe, but I think I'll keep it!)  Perhaps seeing the NoneType error is a timing issue, given that the problem is thread safety.

So, my apologies for the noise.  I guess I'm too used to threads, where it is typical to call the close or stop method from another thread.