Currently exceptions from managed code can take down the native threads, e.g. signalling threads, and this may not be discovered until a signalling API is called.
The native threads shouldn't fallover on managed exceptions.
There is already a try/catch in the managed thread worker around the thread wrappers Run call.