It's in the video, like immediately after the timestamp that I linked. I also provided a direct comparison between Wraith Prism and the L12S so your complaints are moot.
OK, so I've finally watched the video.
1) I'm not saying that guy isn't good at building mITX PCs (I've seen some of his videos earlier), but you shouldn't treat him as a thermodynamics or chip expert. It's pretty obvious this is not his comfort zone.
The first thing that struck me was probably when he says that AMD managed to limit "the amount of power that is lost in the form of heat".
2) The test is performed on open bench, effectively lowering temperatures by a few *C for high air flow scenarios and more for typical systems.
3) First (6:40) he checks how setting power target affects frequency and he measures that:
- power draw of 141W results in clocks of 4004MHz in the benchmark he uses
- power draw of 123W lowers clocks to by 3913MHz
- power draw of 87W lower clocks to 3511MHz
Then (10:30) he shows results of particular coolers:
- L12S: 72*C, 3946MHz, 1750rpm
- 92mm AiO: 76*C, 3904MHz, 2000rpm
- C7G: 87*C, 3829MHz, 2700rpm
- L9a: 89*C, 3802MHz, 2000rpm
Assuming the frequencies are consistent (it's the same benchmark) that means 3950X probably pulls around:
- L12S: 130W
- 92mm AiO: 120W
- C7G: 115W
- L9a: 110W
Which are the data points you should be thinking about, while you seem very fixed on 142W limit.

These leads us to a question: what frequencies would we see under the Wraith Prism?
Because you say Prism is rated for 142W. Well, maybe on a 3900X it is. But based on the above, it's probably more like 120W on the 3950X. And that means you're loosing a lot of performance boost over the 3900X.
And if you want 3950X to fully spread wings in normal conditions (in a case, maybe in a temperature slightly higher than 20*C ambient most reviews meet) you'll have to provide a much more robust cooler. AMD suggests 280mm AiO. I believe them.