Trying to make the best out of the argument, I think you could put it this way: discrete, identifiable vector-like shapes with sharp edges (not only straight rectangles but also rotated ones and other shapes - he also emphasizes typefaces) do require a sufficient pixel density to look good. Skeumorphic photo-like bitmaps fool the eye by hiding the low pixel density using low contrasts between pixels (with natural looking gradients and anti-aliasing). I'm reminded of this when I look at classic 2D video game graphics from the nineties, they make this style really apparent. If it was only about flat rectangles, I'd agree. But I also think this seems to have really started with mobile Windows and the flat single-color rectangles.