It would be interesting, memory willing, to introduce statistically-based random jittering of reflection rays so as to simulate different surface properties. To minimize the noise in the resulting image, each pixel would have to be oversampled several dozen times; since we are not compute-bound on the renderers, this may be possible to do at a reasonable rate. Some problems may arise in keeping this averaged information in memory, and also in distributing random values to the processors in a way that does not reveal renderer boundaries as artifacts. A successful implementation of such path tracing might yield fairly decend and fast global illumination for simple scenes.