There’s actually an interesting tool in Rhino that can analyse what’s going on here. Because there’s a single light in the image, the graph of the value of the color across mostly flat areas will closely resemble the actual geometry. So I loaded up the image into the Texture Palette, and opened the “Graph” section. By fiddling with the V section value, I can get a value graph through the image.
As you can see, there are no dips in in the graph where the dips appear on the image. The flat section in the middle corresponds with the flat part of the model. The dips you’re seeing in the rendering are actually the way the eye reacts to sharp changes in the curvature.
So…it is, as many people have already said, an optical illusion.