If you were creating a skeuomorphic design from simple graphics and then applying filters, textures, and shadows to it via the CPU, the flat design (especially without gradients) would certainly be faster. However, if you properly use a GPU, defining those graphics as textures, allowing it to draw shadows and apply filters, the speed difference would be minute (because it's hardware accelerated and those component would just go idle if you didn't use them -- you could make an argument for power efficiency via power-gating). If you further optimized by applying static skeuomorphic graphics and only had the GPU do basic shadowing, the speed difference would be immeasurable. You can define whatever textures you want, flat or skeuomorphic and the GPU is none-the-wiser. So for you to make a blanket statement like what I've quoted you on, is completely incorrect. I can guarantee you that the differences in performance that you observed were due to changes in your rendering pipeline, not the design itself.
It's like saying, "we moved to text on our poster instead of pictures, because the text is lighter". The fact is that virtually all of the weight is taken up by the poster itself, regardless of what's on the canvas.