Time domain effects might have an influence on listener preference, too. But i would actually suspect that the 2402 horn is better in that respct than the later designs due to no folding / slotting / etc. after the initial compression*). The 2403 could be as good but is second in my book since it sports the same phase plug as the 2405-fram versions of the 2404, so some extra complexity for the sound path.
Back to the frequency domain effects -- i fully concur that the 15 kHz behavior is likely of secondary importance compared to the 9 kHz aberrations. So big Q is, are these dips due to the 2402 horn or are they a consequence of the much-larger 2402 fram? 1Audohack's post #195 showed Bode plots of the 2402H-5 vs the normal 2402H version, but doesn't resolve the issue coz both of his tweeters do not seem to suffer much output loss at 8..10 kHz (the larger fram just a little bit at 8 kHz, but far away not only in severity from the 9.5 kHz dips in MrWidget's post that started this thread).
You mentioned the dip of 2403 to be almost as bad as 2402 in this respect and Grumpy's post #150 points to a possible source, as off-axis responses show a relative level increase for that frequency range, meaning in turn that the on-axis response might be weak at same frequency if we assume relatively flat power response into the room. That said, Grumpy's findings are more in the area around 8, not 9.5 kHz ...
The weekend's coming, will be interested how the audition turns out -- Ralph
*) anyone remembers Fostex 725 that did away with the initial compression?