This was the listening window measured at NRC. Don't you think +3dB from 2-5kHz of the direct sound represents a speaker that's far from "state-of-the-art"? It
gets even worse off-axis.
Moreover, this is corroborated by Stereophile's data
Note the peak at 2kHz again (classic directivity mismatch), matching the NRC graphs, though the scale on the Stereophile graphs make it seem more benign than it actually is. In contrast, the Technics I mentioned has seamless dispersion off-axis throughout the crossover region at 2kHz:
The top-octave peaks are actually artefacts arising from using off-axis curves normalised to the on-axis curve. Dips appear on-axis, but fill in to flat right away off-axis. The normalisation represents this filling in as a peak. It is psychoacoustically-innocuous, and using a measurement method like spatial averaging of the measuring window would be more representative, greatly reducing the dip. Earl Geddes explains
it in the discussion on page 724 of the linked thread.
This discussion on directivity is not theoretical. It relates to sound in-room. As
Toole writes:
What is perplexing for me is a surprising lack of evidence used to substantiate the defense of the LS50 - especially on ASR of all places - and that people whose listening experience have been enhanced by relying on some very sophisticated audio engineering ironically cast aspersions upon using the same parameters their equipment excel in to evaluate cheaper equipment.