In many contests a subset of contestants is granted preferential treatment which is presumably intended to be advantageous. Examples include affirmative action and biased procurement policies. In this paper, however, I show that some of the supposed beneficiaries may in fact become worse off when the favored group is diverse. The reason is that the other favored contestants become more aggressive, which may outweigh the advantage that is gained over contestants who do not receive preferential treatment. Likewise, a contestant may be made better off when a subset of his competitors is granted preferential treatment. The contest is modelled as an incomplete-information all-pay auction in which contestants have heterogenous and non-linear cost functions. Incomplete information is crucial for the results.