Graphic cards performance increase and fast Internet connections are popularising Networked Virtual Environments. This immature paradigm of real-time applications has still to solve a set of internal problems associated with the heterogeneity in client-side hardware, network bandwidth and graphic application requirements efficiently. The problem of properly assigning clients (3D avatars) to servers in a server-network architecture is one of them. This paper describes a new solution to this NPcomplete task, called partitioning problem. This solution is based on GRASP, a multi-start metaheuristic for combinatorial problems. The results of the performed experiments, compared to the main reference in this field for a proof-of-concept system, show better clustering solutions with less execution times.