Small differences in the means of two normal distributions still result in huge differences at the tails. Suppose the top 10% of SAT math scorers make the best engineers. This population will be overwhelmingly male even if there is only a modest difference in the mean scores.