A frequent use case is to scatter dead leaves or small rubbish. Which tend to cluster in piles as opposed to evenly covering the area. So far the best way to approach this is to do pure random sampling with high density, and then manually create higher density spots with small radius paint scope. This is suboptimal, because requires a lot of hidden instances and a complex paint scope. A dedicated fractal sampling would make this much easier and efficient.
I imagine this distribution as an opposite to the random uniform mode, where the more points there are the higher the probability of sampling another point.