I wonder if you measured them individually and then measured them as their sound mixes with the others.
Now I know that sound that returns to you in an echo or a"backwash" through the speaker that is different than the primary projected sound by less than about 30 millisecond is not recognized by the brain. But I still contend that while not recognized as a significantly different tone, it does add to the muddiness of the speaker sound, something that is simply too subtle to show up on a FR curve, but clearly audible in an AB comparison.
As you describe above, real world interactions of sound often are different than simple graphs.