In this work we analyze the characteristics of service overlay networks generated by uncoordinated service providers that deploy different service replicas on overlay nodes across the Internet. Our approach differs from previous works, that generally rely on application-level routing, in that we allow nodes to autonomously re-wire the service overlay to make it capable of absorbing a heterogeneously distributed workload that would otherwise result in some nodes with a specific service being overloaded and others remaining idle. We provide a game theoretic model of the overlay creation process and propose several optimization methods to achieve Nash equilibrium topologies. Equilibrium overlays are characterized by interconnected clusters of nodes that instantiate the same service replicas. Hindered by the computational complexity of finding stable wirings, we propose a simple distributed heuristic that allows the study of overlay networks with a realistic size and with several service instances. We show the ability of our re-wiring strategy to promote the emergence of a clustered global topology whilst running locally. We also argue that the lack of incentives for nodes to participate in the overlay creation might lead to several types of misbehavior, of which some representative cases are analyzed. Finally, both scalability and diversity (in terms of service instances) issues that might affect our distributed heuristic are evaluated in detail.