My remark was actually aimed at winegstreamer and other libraries that create Unix threads. We should avoid that as far as possible.
Is there a reason for this besides killing threads not (currently) working?
As far as avoiding the crashes, it seems to me that now that the syscall interface is in place it should be possible to properly unwind the kernel stack.
We can't avoid GStreamer creating threads.
I don't think that GStreamer is the problem behind bug 52213, though. The crashes I've seen in practice have all been from winealsa (or other sound libraries?), and I have some patches that should let those threads exit cleanly. I haven't sent them yet since the mmdevapi rework is still in progress.