In studying the dynamic behavior of processes in artificial or natural social systems, a key factor is the topology of the social network structure. It has been shown that real-world social networks tend to have non-random network structure with properties such as short average path length, excess clustering, and skewed degree distributions. We show in this paper that the structural nature of the social network of artificial or simulated agents has a significant effect on the performance of the overall system. We conclude that finding "good" network structures for a particular application domain is critical to modeling artificial social systems and implementing multi-agent systems. We argue that techniques for adapting network structure will be critical in large-scale agent communities.