Distributed search and track is a canonical task for multi-robot systems, encompassing applications from environmental monitoring to disaster response to surveillance. In many situations, the distribution of objects in a search area is irregular, with some areas having high object densities while other areas have low densities. In this paper, we formulate the search task as a multi-armed bandit problem and propose a novel distributed formulation of Bernoulli Thompson sampling that enables robots to share coarse global information across the team. We demonstrate that this new formulation significantly accelerates the speed at which robots find targets compared to previous distributed search approaches. This effect is even more pronounced when the distribution of targets is clustered within small subregions of the search space.