Despite recent advances in rendering hardware, large and complex virtual environments cannot be displayed with a sufficiently high frame rate, because of limitations in the available rendering performance. This paper presents a new approach of software accelerated rendering which draws from the concepts of impostors, hierarchical scene subdivision and levels of detail. So far software optimization in real-time rendering has merely considered individual objects. This work is actually optimizing the rendering of the whole virtual environment by implementing a three dimensional image cache. It speeds up rendering for large portions of the scene by exploiting the coherence inherent in any smooth frame sequence. The implementation of the three dimensional image cache is discussed and the savings in rendering load achievable on a suitable hardware platform are presented.