I still would not rely on that optimisation, though. If I think this could be an actual bottle neck, I would make the shared buffer explicit in my code.