Nah, you can show the on-axis response separately.
If it was a bad frequency response, It is SO OBVIOUS, you can not hide it.
And if it has truly good directivity, there is absolutely no problem with normalizing, which KEF is not in the case.
You are listening to the direct path at "some-axis", not the average of LW, so unevenness of the on and off-axis response inside the LW is already a bad thing.
View attachment 187680
Well, 1/3 octave smoothing is absolutely SH*T, I agree with you, but where comes the SH*T from at the first place?
Miss-quote maybe?