I think that a lot of these subs have the opposite of what you are thinking. A lot of micro subs now will play as low as they dare, even down into the teens hz, but will reduce low frequency output as SPL increases. This way, they are only flat when playing at 75-80 dB. The dynamic DSP knows how much power the amp has and how much excursion the driver has and throttles back the bass accordingly at higher SPLs. DSP makes it so a tiny box can play to sub 20 hz or can play to 100+ dB, but physics dictates both can't happen at the same time.
For a studio monitor sub like he Neumann, the designers would much rather the sub go into protection or distortion rather than exhibiting a non-flat response.
I am not thinking, I'm just reading the results.
These non linearities are perfectly explicit in the tables :
Consider a signal made of two tones of equal amplitude at 40 and 50 Hz.
With 0.2V at the input of the B&W sub you get almost the same amplification at the 2 frequencies (0.5dB difference) , but at 0.5V the difference is much larger (almost 2dB).
B&W
0.2V => 98.68 dB@40 Hz / 99.17 dB@50 Hz (difference = 0,49 dB)
0.5V => 105.14 dB @40 Hz / 107.07 dB @50 Hz (difference = 1.93 dB)
So you see a non-harmonic distortion of almost 1.5 dB, but that's really a pity, because the subwoofer CAN cleanly output 107 dB @ 40 Hz, but this requires 1 V at the input.
With a clever processing the sub could have been transparent for these 2 tones up to 107 dB (or a little less) while in this case the difference exceeds 1 dB as soon as 104 dB.
I think this is called compression.
What I observe is that the Neumann optimizes its physical capabilities : the problem happens only when the device reaches its physical limits while in the B&W the problem appears progressively over a much larger input range.
The Neumann dsp seems to compensate the non linearities of the analog transduction while there is apparently nothing of that sort in the B&W.
My questions are whether any convenient figure or chart quantifies these problems.
I also wonder how important this is compared to the usual linear harmonic distortion.
1 or 2 dB may not be very important, especially at those low frequencies (?)
When Amir tests a speaker he publishes the distortion at multiple output levels.
It would be also very interesting to compare the frequency responses at those different levels.
We could normalize at, say, 1 kHz and draw the relative curves. A perfectly linear speaker would show exactly the same curves, even if not perfectly flat.
If the curves are not identical then a parametric EQ correction could be fine at a certain level and bad at other levels.