Foveated 3D Graphics (2012) [pdf]

I always thought this was an approach that would be well suited to ray tracing or similar. Both ray tracing and rasterisation are proportional to scene complexity and pixel count, but the balance is different. Ray tracing handles scene complexity better and rasterization covers pixels quicker.

The excellent http://www.adriancourreges.com/blog/2015/11/02/gta-v-graphic... shows how many pixels are written to to construct a scene using rasterization. Many many intermediate render targets are involved. Any intermediate layers that are not directly in screen-space would not benefit from foveation.

Raytracing/ray marching etc. working on a pixel by pixel basis would be far more suited to larger fuzzy pixels where you are not looking.

This is definitely the future for both desktop and VR gaming. There is no way games will go to 4k and beyond at 2x60hz and spend 60 rendering passes doing pixels in the top left corner of the screen. The question is what will happen to console gaming? The eye tracking issue isn't quite as easily solved for consoles.