Not quite. It’s more like (still simplified):
vertex shaders
triangle setup
pre-raster
pixel shaders <= texture units
rop
The Pascal multi-viewport feature saves on vertex shaders and (IIRC) triangle setup. Not on anything after that. Note that texture is a kind of adjunct to pixel shaders–it happens at the same time. Every pixel that access a texture requires one texture sample (or more). It’s basically a fixed ratio.
That said, I did oversimplify and this leads into your FOV question. Even at ~110 degrees, there is a high amount of distortion off-axis. It gets even worse with higher FOVs.
Multi-viewport can help not just by rendering both eyes at once, but also by rendering multiple times into each eye. Furthermore, each viewport can be a different resolution. You can thus lower the resolution at the edges of your screen, since it is distorted anyway, with no significant quality loss.
This multiresolution technique saves both shader and texture horsepower, since you render fewer pixels on average.
Note that while you could have done this on Maxwell on earlier, it would have required submitting the geometry many times, like 8x. That’s obviously very significant and makes the technique only practical on Pascal.