Distributed virtual environments, which simulate an actual physical or imaginary world on a network and allow multiple participants to interact simultaneously with one another within it, are becoming increasingly important for both research and practical purposes. As the number of participants and the amount of information exchanged among participants increase, it is crucial to large-scale distributed virtual environments to overcome bandwidth limitations and resolve network latency and synchronization problems. We present a new framework, called MELD, for modeling distributed virtual environments using the pilot/drone paradigm, which allows each host to locally model remote entities in order to resolve latency problems and improve responsiveness. Our approach uses shared event queues and a cache coherence protocol to synchronize the pilot/drone. To further improve the system's scalability, interest management is used to filter unneeded data before a host receives it for processing. The partition, however, introduces the problem of dynamically joining a group in a real-time simulation. We address this problem by presenting a checkpoint/restart mechanism based on an action hierarchy and a parallel finite-state machine structure. Additionally, ALOD (action level of detail) is employed to mitigate the lag between pilot and drone at any joining time.