Olympic Timekeeping: How small a difference really matters?

The 2004 Summer Olympics is a technological tour-de-force. Events are timed to the thousandth of a second. As broached in this article, how fine a time grid is fine enough? Does a thousandth of a second difference between two runners, say, really reflect a difference in performance? Does a hundredth of a second difference reflect a real performance difference? Just because we can measure smaller time slivers of differential in human performance, does it mean that we should?