The <40ms CSD quote probably comes from this paper:
https://www.semanticscholar.org/paper/The-Yamaha-NS10M:-twenty-years-a-reference-monitor.-Newell-Holland/137bc09b853ae8ae4392e24e8c4b93ec523aeb98
The Waterfall of the NS10 is indeed rather clean but you can take any other decent speaker with trouble-free midrange (when ported then with rear port unless the front port is well damped) and EQ it to the same target and the CSD looks the same. For example, #35 in the paper, the Westlake BBSM-5. A ported one obviously needs to go much lower than the NS-10 in order to be able to EQ it to the 12dB slope for long enough.
Vice versa, when you give the NS10 the response of a ported speaker by adding a 12dB high-pass (subsonic filter) it will look exactly like a true ported one. Often, ported active monitors have subsonic filters as well, some 2nd order, some 4th, yielding 6th and 8th order roll-offs which of course are "slow". Only way this doesn't have much impact is to move the roll-off to 30Hz and preferably even lower. A ported bookshelf mini-monitor with a 50Hz corner and 4th-order subsonic, so 8th order total (like the A5X), will always have lagging "weight" on bassdrums etc.
After all frequency response (mag and phase) <--> transient response still hold and will always hold. Linear Systems Theory cannot be cheated on.
As for CSD Waterfalls, a CSD is just slices of FFT's over time of the impulse with short windows (hence low resolution in the bass). The lower the level in the bass the faster the CSD drops below the "floor" in the plot and lower the order of the bass roll-off the more so as the ridge decays faster.
Comparative decay statements must be made with the passband response normalized (based on 1/2 or 1/1 Oct smoothing so as not to overcorrect the fine grain).
There are better methods to visualize the decay than old-fashioned MLSSA-style CSD. Frequency dependent windowing should be used for the slices and the time axis should be periods. And of course, as mentioned, made on the trend-line normalized response, not the raw response.