Achieving an efficient global illumination method is one of the more important current issues in computer graphics. The main problem is that the simulation for large scenes is a highly time-consuming process. In this paper we present a new non-uniform partitioning method in order to achieve a good load balance on a distributed memory system. This method is based on the minimization of a distribution function. Furthermore, we have developed a scheduling based on processor hierarchy which permits the utilization of a visibility mask with crossed frontiers. Finally, in order to evaluate the proposed method, we have used a progressive radiosity algorithm, running on a PC cluster with modern processor and network technology as a testbed. An efficient performance in terms of execution time and speedup has been obtained.