The owner thread gets terminated while every known handle for the owned mutex have been closed already, but while a thread is still waiting on it. From wineserver perspective the mutex doesn't exist anymore, and it has no way to know about the remaining waiters or owners, and this would need to be resolved on the kernel side.
Well, technically we could deal with it ourselves, by not losing track of the fd. An early revision which I don't think was ever submitted did this, by making inproc_sync a full object with a handle.
But it's also not the only thing we'll need to fix, so it's probably at least worth seeing if we can get a new API for this.