Ideally we'd make the reviewers work harder by insisting on various resolutions and settings even for CPU reviews.
Some of the higher quality settings do use CPU resources so running at 640x480 or 800x600 in low quality settings doesn't necessarily always show how well a CPU can cope with some theoretical monster drawcall load in the future.
Another thing reviewers do not cover is having any kind of streaming going on (ShadowPlay, ReLive etc.).
Also, multiplayer is very tricky to benchmark. This goes back to BF4 and Mantle, where benchmarks noted only small differences but yet people playing multiplayer noticed huge differences with smoothness.
Someone over on the AT forums posted this: