Actually that makes sense. I have found one of the issues when working with coaxial speakers is differences in the on axis and off axis response die to cone reflections from the tweeter output causing peaks and nulls. So yes, it is not surprising to me the off axis response would be best.
Paul Klipsch published a research paper on this years ago, I think in AES, where he investigated all this, and also the doppler distortion that arises in these reflection due to cone motion of the bass driver.