It seems objectively correct, though. The model is simpler. They might take the model and do more processor intensive work with it, transforming it, tessellating if, etc. before render (like all models). But the model itself, is simpler.
Except that you can't separate model and normal maps since they're baked versions of the real model. As, you can say you have smaller textures (but shaders with complex 2D data also)
