No, I understand it all quite well.
And no, the edges will not look the exact same. It's not an exact culling of all wasted pixels. You reduce the penalty of *some* of the culled pixels, but much of the image is still there and without supersampling, the outside edges WILL look blurry.
You could do a very minor % with the multireprojection to keep the edges looking as clear as possible(lenses obviously still dont help), but you also minimize the performance gain from it. The more extreme you go, the more you're affecting visible parts of the render output and the more performance you get from it. I would guess that in the example used, they are rendering some 50% of the pre-buffer render at a lower resolution(at what % of lower resolution they dont say, which also matters) to get the sort of gains they are claiming. And this will ABSOLUTELY affect a noticeable part of your usable FoV.