Less cynically, it also increases the odds of finding good employees. A "bottom 5%" worker is statistically less likely to suddenly rise to the top of the pack than a new hire is to produce at that level initially. And, more cynically, that remains true even if there's significant measurement error in identifying that 5%.