The trouble with simply rendering the original render target at double resolution and distorting that straight to the screen is that this would give the same effect as using a texture that has no mipmaps and is too high a resolution for the target area. It would just be "dotty" and not correctly antialiased. In fact the antialiasing effect would be of varying quality across the image. In some small local areas near the eye centre, the scaling and position would be just right to interpolate 4 texels to give the intended resulting antialiased pixel. Just a little way from that area the scaling and position would be exactly in the wrong place and the output pixels would take their colour from alternate texels, missing out any data from the intermediate texels. Moving out from the centre it would get more and more dotty as the destination area becomes smaller and smaller, so the source texture's resolution is more and more too high. Near the corners it would just be really dotty, a lot worse even than the current version which already demonstrates (in the corners) the problem with rendering at too high a resolution.
The two stage option is to guarantee correct antialiasing on the first stage, because the position and scale are chosen so that the intermediate texels are always a perfect combination of 4 original texels. I don't really know how good the result of that would be. But the first thing I will do is make LFS choose a render target size that is "correct" (1 texel to 1 pixel scale) in the centre of the views, whatever output resolution you select. This will allow downsampling to perform some antialiasing on the final image and should help a bit. Also this simple update allows higher res Rifts to be used if you work at Oculus VR.