Maybe there's still merit to BD's approach. The rule is: first make it work, then make it good, then make it fast (i.e. optimize). It's probably easier to find plenty of ways to incrementally improve on a working platform than to dream up those improvements completely from scratch.