It seems like this would be a bit difficult to implement in our raytracer since we are simulating multiple bounces. I suppose you could use disk sampling for the first bounce, but after that all the bounces are from different locations (and may have different BSDFs).