I think it’s you that has failed to grasp what the problem is here. Games perform differently with different drivers, Steve himself showcased this not that long ago, showing the Nvidia driver to have a higher CPU overhead than the Radeon driver. This means that we would likely see bigger differences between these CPUs if you used an Nvidia card. And as like 90% of the market uses Nvidia it would have given better results to use Nvidia.
It’s very simple to understand.
You're creating an idea in you're head that doesn't equate well to what you're trying to equate it to. He's testing one CPU against another CPU. An ideal test setup is the one that gets rid of other constraints, because ANY constraint in the system will act to slow down the total system. You want the CPU to run with minimal constraints from the rest of the system.
From all the reviews I watched about Ampere and RDNA 2, the best of these, when put up against each other, when game testing @ 1080p the 6900 XT gives the best results. Therefore, testing @1080p and using the 6900 XT gives the least constraint from the graphics subsystem. Even if you run Nvidia though, running at 1080p gives the least constraint from the graphics subsystem and gives you the biggest delta to make comparisons with. He also FINALLY got out of the stupid notion that testing with 3200MHz memory @ CL14 doesn't effect the delta for CPU testing, when clearly other reviewers have shown it's a constraint. So, he FINALLY put in the correct memory to run with these newer systems and stepped up to 3800MHz CL14 memory. It benefits both Intel and AMD, but for AMD, it allows for the interconnect to run at about its peak performance, while running at 3200MHz slows it down. It's a difference of running the interconnect at 1600MHz or 1900MHz, and anyone with a brain could tell you that's going to make a difference, especially if you're trying to reduce latency, or, remove constraints from a test setup.
Why it is you feel it would be better to put in a GPU that would run slower, which there is SO MUCH historical data in testing that says you get less of a delta when you do that, is beyond me.
Steve's test setup is now ideal for all current gen hardware, although a Samsung 980 Pro thrown in as the NVMe drive would once again, reduce constraints. But, it has no bearing on frame rates in any games he tested. If he were doing a complete set of productivity tests though drives can start to affect numbers. It wouldn't be the case in most testing, but at some point it would make a difference.
BTW 90% of consumer don't buy the best Nvidia GPU that exists on the market, and Nvidia doesn't sell 90% of what consumers buy when it comes to discrete GPUs that get put into systems. Their numbers show large because data that's often used to represent market share comes from Steam, or at least what Nvidia likes to show. The problem with that is they don't exclude countries, and in many countries most people don't have a PC, and their using a PC at an IC to play games. I could walk around the many ICs where my wife is from and take PICs to show how that works. But, Steam data is based on user accounts, and if in Asia and Africa and S. America most machines have an inexpensive Nvidia GPU, then that skews the data since each user of the same machine shows as an instance of a GPU. In other words Nvidia shows bogus data for market share. Based on sales data AMD has a lot more than 10% of the market over the last few years.
But if you want to test based on what most people own, then you'd need to test with a low end GPU, which would give you about a ZERO delta with the new gen CPUs.