And remember that, practically, when you sum up multiple overlapping gaussians you still just get a single gaussian in the end.
So multiple "layers" of shadows seems to be entirely redundant and simply a waste of computing power.
(Unless you want 2+ shadows because you're trying to simulate the effect of 2+ directional spotlights with hard edges, but I've never seen anybody ever want to do that in a UX context. Modern UX shadows are almost universally simulating a single softbox light from slightly above.)