This was closer to Gavin's orginal suggestion too. The (an) algorithm isnt going to be perfect, but its going to be better than repeated hard forks (in that it may or may not require repeated hard forks). I'd prefer something that adjusts so we can have a little more time to as we scale up.