At first glance, shared address space seems to be much faster than message passing. But in massage-passing, we know all synchronization points. Couldn't that enable a lot of optimizations?