Will something like `Arc<Mutex<T>>` be optimized?

That's the part of hightly debated question right now.