A detailed longitudinal dataset is assembled containing annual performance and biographical data for every player over the entire history of professional major league baseball. The data are then aggregated to the team level for the period 1920-2009 in order to test whether teams built on a more even distribution of observed talent perform better than those teams with a mixture of highly able and less able players. The dependent variable used in the regressions is the percentage of games a team wins each season. We find that conditioning on average player ability, dispersion of both batting and pitching talent displays an optimal degree of inequality, in that teams with too high or too low a spread in player ability perform worse than teams with a more balanced distribution of offensive and defensive talent. These findings have potentially important applications both inside and outside the sporting world.