Complex model samples very slowly - wondering about possible improvements

I think @twiecki or @aseyboldt will be able to give some pointers here