On Win32, run the attached test case twice, in two different VMs. Try to drag
the button from one VM to the other. The VM on which you drop appears to
complete the data transfer, but then attempts to *exit*. At this point, the
VM hangs because the ShutdownHook code in WToolkit cannot execute normally.
Truly bizarre. The bug only happens when you transfer a serializable object
across VMs.

The Win32 native message pump which keeps the Toolkit thread alive must be
exiting. This could happen if we posted too many WM_QUIT messages to that
thread. But all of the calls to
startSecondardEventLoop/quitSecondaryEventLoop
and MessageLoop/QuitMessageLoop appear to be balanced. Alternately, we could
be raising a SIGTERM signal, but I don't see how that could be possible.