I wonder if to get closer to the true answer, we can take multiple Monte Carlo shadows and then layer them on top of one another? And I guess somehow we'd have to make sure that the color stays true (that it wouldn't become too dark?).
edit: Just saw the following slides, I guess my intuition was correct.
To add to the above comment, it seems to me that we can achieve a similar result by increasing the number of samples.
When Ren mentions adding more samples this means that when we sample on the surface at a pixel location, we take multiple rays at different directions towards the light source?
Yes, since each sample is sampling a ray from a given pixel towards a random direction (or towards light source when using important lighting). Increasing # of samples would be taking multiple such rays at multiple random directions.