The difference here is that, the 3900x and the 3950x lose in far more single/lightly threaded scenarios, and not only to the 10900k, but even 6 core and 8 core Intel chips! In other words, your specific workload must be throughput-oriented, and run better on AMD to justify getting the HEDT R9 chips. That's called niche.
Are you still talking about the 10900k v 3900x v 3950x?
Your specific workload must be throughput-oriented and run better on 10 cores to justify buying the 10900K over a 10700K, right? But that's not the case with the vast majority of desktop user workloads when you're trying to label the 10900K as a value proposition, or a price to performance proposition. Hence, the 10900K is a poor value proposition for most desktop users because they'll rarely need 10 cores/20 threads over 8 cores/16 threads (or even really 6 cores/12 threads). The 10900K, just like the 3900X and 3950X, is for niche cases. Again, it's a bad value proposition, just like the 3900X and 3950X, for most users.
With respect to overall single-lightly threaded scenarios, if you take Puget Systems' test suite (largely lightly-threaded as it is based on Adobe's suite (and I'll even remove CB20 MT from that result and replace it with AnandTech's ST CB20 score) and add in Anandtech benchmarks that are lightly threaded, like y-cruncher ST, Dolphin, GIMP, Digicortex, 3DPM ST, x264, GB4 ST, and all the gaming at 720p/768p, you get an average lead by the 10900K of 9.89% over the 3950X.
In the multithreaded tests like POV-Ray, Blender, Corona, 7-zip combined, y-cruncher MT, 3DPM with AVX, CB20MT, you get the 3950X out on top by 27%.
There are compromises no matter what chip you pick.
And if your argument is that few desktop users are using Blender/Corona, 3DPM, Cinema4D (heavily multithreaded), the same could be said of Agisoft Metashape, Digicortex, Dolphin, x264 HD 3.0, and Far Cry 5 at 720p (very lightly threaded or single threaded).
So again, the argument that there is any one single "best" chip depends ONLY on what your use case is. Period.
And if you simply average out the single/lightly threaded 10900K lead of 9.89% with the MT lead of 3950X you get the 3950X ahead by over 15%. Truly DOMINATION by the 10900K right?