Very interesting! Myself, I use the anisotropic node with turbulence pluged into the bump channel. But your comparison shows me that I should experiment with blinn for situations with softer lightning.