Favoritism towards a relatively weaker competitor is widely adopted as an effective instrument to enhance productive effort provision in asymmetric competitions. In this paper, we investigate the effort-maximizing favoritism rule in asymmetric two-player contests with all-pay auction technology, while accommodating fully flexible (nonlinear) favoritism rules. We assume that the players’ competencies (measured by their values of winning the competition, or marginal effort costs) are public information. We find that at the optimum, the weaker player is extremely favored; however his/her winning chance converges to zero. This finding illustrates that the effort-maximizing extreme favoritism rule perversely decreases winner diversity.