The assignment of resources or tasks to processors in a distributed or parallel system needs to be done in a fashion that helps to balance the load and scales to large configurations. In an architectural model that distinguishes between local and remote data access, it is important to base these allocation functions on a mechanism that preserves locality and avoids high-latency remote references. This paper explores performance considerations affecting the design of such a mechanism, the Concurrent Pools data structure. We evaluate the effectiveness of three different implementations of concurrent pools under a variety of stressful workloads. Our experiments expose several interesting effects with strong implications for practical concurrent pool algorithms.