If it were more than 100, would sampling solid angle eventually match the right picture?
From my understanding the slides in the beginning of this presentation was testing the blocker's affect so we test directions we know are towards the light source, but now it is any randomly sampled direction?
it's amazing to see how much the importance sampling improves the integral value. Since we know what objects emits lights, it's fairly simple to get a desired probability function.
Is there any merit to performing solid angle sampling and storing all the ray bounces in some sort of data structure? This could possibly enable adding and moving lights in the scene with greater ease, as rays wouldn't need to be recalculated but rather referred to in a look-up table. Sounds a little crazy...