To support large-scale users to share a virtual environment, multi-server architecture is applied to the virtual environment system and each server handles a partition of the virtual environment. The unpredictable movements and interactions of avatars may cause some servers to be overloaded. Existing load balancing algorithms try to redistribute load among servers and incur too much overhead, which may reduce the interactivity of system. In this paper, a layered iterative dynamic load balancing algorithm is proposed. Setting the overloaded region as the center, the algorithm chooses a limited number of neighboring regions as the load adjusting goals and iteratively spreads the overload outward from the center. The load balancing state can be achieved through iterations. Based on the virtual environment wherein avatars are scattered under skewed and clustered distribution, the algorithm is built and comparison with the existing load balancing algorithms are made. The results show that this algorithm can adjust load efficiently and incur less overhead.