Stochastic simulations of complex systems from domains including physics, biology, ecology or economics often require large system sizes, long time scales, and numerous replications to fully explore model behavior. The simple rules defining many models can lead researchers to prefer familiar but inefficient programming techniques, which severely hinder progress by creating computational bottlenecks. While such studies often benefit from combined domain-specific, statistical, and programming knowledge, few individual researchers span the full range of necessary skills. Here, we present a collaboration on the neutral model of biodiversity in dendritic river networks, where the goal is to analyze biodiversity data across the world’s major river systems. We show how we achieved large performance gains by engaging the problem at its foundations and thereby enabled research at a new scale.