Jebus, you weren't kidding... that's a lot of distortion. In one example we're talking a visual difference of >50% which amounts to an actual difference of 80

Jebus, you weren't kidding... that's a lot of distortion. In one example we're talking a visual difference of >50% which amounts to an actual difference of 80 "points" out of 10000, so <1%. (No, that's not a typo on my part.) You have to wonder of most of the visual difference isn't actually just statistical noise at that point.