The object-space parallel processing for global illumination models is one of the most promising approaches to fast photo-realistic image synthesis. However, there is a potential bottleneck between processing elements and a frame buffer in massively parallel processing systems based on the object-space parallel processing, and this factor may restrict their scalable performance. To solve this problem, this paper presents a novel frame buffer system, named a cached frame buffer system. By adopting the cached frame buffer system into the object-space parallel processing systems, the overhead of the frame buffer access due to conflicts and long latency can be reduced, and the potential of the object space parallel processing system with a large number of processing elements will be fully exploited.