On top of that, the CPU can literally elide movs at runtime so the whole swap could end up being zero latency depending on the code motion