Coordinating the action of distributed agents in a dynamically changing environment is a difficult task. The difficulty stems from the fact the real-world environment changes independently from the actions of the agents. Furthermore, the agents may have limited competence and awareness about one another's decisions and about the environment. This paper focuses on cooperative plan execution and presents a dynamic threshold-based scheme for balancing load among distributed agents. Different load balancing strategies, namely client-initiated, server-initiated, and hybrid, are discussed, and a comparative analysis of these strategies is presented. The study shows that dynamic threshold-based strategies perform well compared to static threshold-based methods. The study also shows that dynamic threshold-based strategies greatly reduce the communication overhead, and are well suited for cooperative plan execution in dynamic environments.