Maybe some restrictions regarding games anda settings that use more that 4 GB of memory.
Not that I'm agreeing that this happens but if it did, then restricting benchmarks to 4K would push things very much in favour of AMD while 1080p would favour Nvidia. This does account for some of the huge variations we see in performance between Nvidia and AMD. That and the games and settings used. I do wonder at the logic of these selections from some sites. Often a mixture of ancient or fairly obscure games at weird settings. Personally, for high end video cards I'd test everything maxed out (minus Nvidia exclusive Gameworks features or features which have an extreme bias towards and are sponsored by one particular vendor (hair works)) apart from MSAA at 1080p, 1440p and 4K in all of the games below and include a performance summary/average across all the games at each resolution. That's about as unbiased as you can get IMO. These games have been selected purely on the graphical requirements and popularity while trying to get a mix of genres in there.
GTAV
Watch Dogs
AC: Unity
MGS5: Ground Zero's
The Evil Within
Ryse
Project Cars
The Witcher 3
Dragon Age Inquisition
Farcry 4
Crysis 3
Wolfenstien Old Blood
Battlefield 4/Hardline
Call of Duty Advanced Warfare
Metro Redux
MGS5: Ground Zero's
Middle Earth: SOM