There are a variety of choices which need to be made when setting up a multi-agent community. In particular, which agents communicate with which, what protocols they use, and what information flows from one to another. Such design choices will affect the efficiency of the community with respect to several parameters - accuracy, speed of solution, and message load. In this paper, we consider one class of problem which multi-agent systems engage in - service provision. Using a simple, abstract, form of this problem, we use a mathematical analysis to show that three different messaging protocols result in varying message loads, depending on certain parameters such as number of agents and frequency of request. If the parameters are fixed, we can conclude which of these three protocols is most efficient. However, these parameters will usually vary over time, and hence the best of the three protocols will vary. We show that the community can adopt the best protocol if each individual agent makes a local decision based on which protocol will minimise its own message load. Hence, local decisions lead to globally good behaviour. We demonstrate this both mathematically and experimentally.