Performance pay schemes in public bureaus may lead to dysfunctional behavioral results, such as gaming. In the aggregate, unintended outcomes may arise. Interestingly, gaming effects may not be uniform across all bureaucratic agents when agents are differentially motivated. I investigate for whom this behavioral result arises. To do this, I use a two-wave stated preference extra-laboratory experiment where, after completing a pre-experiment questionnaire, subjects are randomly assigned to one of two simulated public welfare organizations – a control organization and a pay-for-performance organization – with identical objectives and asked to make a series of choices regarding effort allocation to hypothetical clients with randomly assigned attributes. I present three main results from the experiment. First, there is some evidence for gaming overall. Specifically, I find that subjects shift effort away from the least employable clients and towards the most employable clients in the performance pay organization. Second, while preferences differ markedly, I find that the magnitude of the gaming effect is very similar for subjects low and high in public service motivation (PSM). Third, at the extremes, client outcomes change little due to preferences for helping the least advantaged, although the difference in support between the most- and least-advantaged drops significantly. Thus, the best predictor of outcomes under a performance pay scheme may simply be agency composition. These results have implications for our understanding of strategic response to bureaucratic incentives and public sector work motivation.