Point density isn't accuracy, why not actually compare it to the lidar data and to the Codemasters version instead?
(Also, it's not 15-20cm, these tracks import at 0.01 scale)
The comparison to rf1 is about point density, which as I said above is not accuracy, but detail.
Plenty of that detail is lost in an rf1/fm7 level of density, at least from what I've seen in most fm7 tracks (long beach was mostly 1 big square every 2-4 meters, which felt like crap in the previous, non-smoothed conversion)
Also,
"I feel much more difference (in positive direction) from a very skillfully created, but not scan-based track (e.g. Reiza tracks) to a 15-20cm scan, than from a 15-20cm to a 1-2cm scan."
Have you had a chance to try rfPro in a specialized sim? Cause otherwise I seriously doubt you've driven a track made out of a 1-2cm scan, definitely not with that point density in the actual mesh.
Rf2 scanned tracks are more or less around 1m-50cm dense, ac and acc look similar tho I've never actually measured it and I doubt iracing would use any more than that.