It's possible that your speakers have a distortion which is present but below the detection threashold at lower levels, and then as you go up to high levels, the distortion becomes louder than the decection threshold. In other words, your ears may have a non-linear perception of what is fundamentally a linear distortion.
It is also possible that your speakers have a distortion or coloration that is level-dependent, or non-linear. For example, if the voice coil of your midwoofer is heating up faster than the voice coil of your tweeter, then your midwoofer will compress more at higher levels. So you could end up with a brighter tonal balance at high levels.
My instinct is that the first situation is the case, this based on my work with horns. In effect a Uni-Q driver horn-loads the tweeter, but much about that "horn" is not optimum from a distortion-minimization standpoint. We see this in prosound with a sub-optimal horn: It sounds fine at low levels, but crank it up and it sounds harsh. The distortion was there all along, but we don't hear it at low levels.
Imo, ime, ymmv, etc.
Duke
dealer/manufacturer