it really seems to me Harman engineers optimised for linear LW, not on-axis.
That might well be true, and I'm not saying that's not a reasonable design philosophy. In fact, if I were in their shoes I would likely do the same. But just because linear LW was their priority doesn't mean that individual angles within the LW are not important. These two things are not mutually exclusive. You can optimize for LW and also try to make sure individual angles within the LW don't deviate too much from the average.