CPU limited when they weren't underclocking the gpu to mimic the Switch. When they underclock the gpu to 570mhz it's the GPU that's running at 100%. Basically CPU, GPU and Memory Bandwidth are all going to be challenging issues when it comes to porting Witcher 3 to Switch.

last 2 paragraph when Richard mention PC to Xbox 360 by reworking the engine, then you back to first 2 and 3 paragraph where Xbox 360 have a powerful GPU that equal to standard or minimal requirement for The Witcher 2 spec and compare it to Tegra GPU that is not even a desktop GPU. 

So last gen reworking engine are made on console with desktop GPU class, while  if they even able spare times to focus on Switch it still be the same result, because tegra X1 is mobile GPU. 

His reasoning is pretty flawed, just like it was for DOOM. The OS takes up resources, Windows using A LOT more than the Switch OS, the performance of his build to how Switch will handle it is not 1:1, and some of these ports devs have reduced geometry and lighting to improve performance. That doesn't mean a port is possible. It just means that DF's reasoning on why it isn't is pretty off-base.

It would be incredibly difficult, the twins already have a hard time running this game in 30 ish FPS. I really wonder how they could port this game to the Switch without practically remaking it.

Still Doom was able to run in 60 FPS on PS4 (best refference, since the PS4 OS is also pretty light). So they could make the Switch 30FPS. Which is a big thing since the power required for 60FPS is pretty big. The PS4 can barely run the Witcher 3 in 30FPS. So expect the visual difference between the Witcher 3 version of Switch and PS4 to be much lager than those of Doom,which where already pretty big. Doesn't mean that it is impossible of course just a huge amount of work. 

Isn't that was I said? They could reduce geometry and lighting. My main point is that this method they used to try to debunk it is inherently flawed in concept.

They mention the impact that windows has on the CPU in the video. Then they switch to a more powerful CPU and show the GPU and Memory Bandwidth is a problem.

It could easily be done, it would just need more work than a bump down in resolution.

All of this completely ignore the difference in architecture, though, which affects performance and the entire graphics pipeline. Just looking at the raw numbers and approximating the benchmarks will NOT give you a clear idea, because it can never be 1:1.

Possible to do? Yes. Easily done? Not at all.

But can give you a ballpark of what cuts may be needed to run it and if it wil be acceptable.