Don't be fooled.
The actual directivity is not that clean, it's NOT normalized.
KEF tend to compensate to the listening window, and the on-axis response has bumps.
The first contour line might be ref SPL-3dB.
And you should remind even a 1dB bump in the on-axis corresponds probably over +5deg in the -3dB beamwidth.
This is the trick they used to make the "appealing figure" for their catalogue.
Genelec 83*1 is a hell of a lot cheaper and the measurements look great without any cheap trick because it actually measured superb.
I think "cheap trick" is unfair. It's true that KEF's Uni-Qs are uneven precisely on-axis, but their reasoning is that it's the listening window that matters more, and that the unevenness is VERY narrow - it's an interference effect from the tangerine waveguide that only occurs precisely on-axis that is only really detectable with measuring devices that line up precisely with it.
I feel that normalising dispersion plots to on-axis frequency response in that case makes them look much worse than they really are. The lumps that become visible are not
real dispersion unevenness, they're just artefacts of the precisely-on-axis unevenness.
If the on-axis-normalised dispersion plot looked smooth while the on-axis was uneven, all that would mean was that the very localised unevenness had spread, which you wouldn't want.
Normalising against listening window response feels like it would be a better approach generally. And arguably normalising dispersion plots
at all is a cheap trick to try to show good dispersion of a bad frequency response. The unnormalised plot is reality - the actual frequency response being dispersed.
(Maybe they could actually sidestep the issue by declaring the reference axis to be a few degrees off centre, as the Spinorama rules permit. Then the on-axis unevenness wouldn't show up at all, being between sampling points).