I included only the corrected part of your graph as that is the one addressed with my comment - single point sweep measurement of uncorrected response is equally not representative.
Spatialy averaged measurement of both, frequency and phase, is needed for a precise room EQ and also for the precise control of the room EQ results. For spatially averaged frequency response you have 2 options with REW - either take a number of sweeps from different points around the listening area you are optmising and than use Average tghe Responses under All SPL to average them or use MMM RTA. If executed properly both will give practically identical results but with MMM you can process more measurements in a less time.
For spatial averaged phase response you have only one optionwith REW - take a number of sweeps from different points around the LP and use vector averaging. Frankly, I have rarely used that as phase response doesn't change so "wildly" from point to point as frequency response might, so a single sweep is usually enough.
And I don't agree that anything has changed from the time Geddes and Blind wrote that article - it is a scientific fact that single sweep contains no spatial information but only informatioin about the response in the particular point in space where mic was located while measuring the sweep. For that reason no magical algorithm could be applied to extract spatial information from a single sweep because spatial information is simply not there. Situation where FR taken from single sweep matches the averaged response from spatial measurement is a pure coincidence related to the particluar room you're doing meaurements in so it's certainly not something that should be repeatable in any given enclosed space.
Unfortunately, the only way to collect spatial data is to take measuremets at multiple points in space.
But unlike your single sweep, most high-end room EQ systems (including Dirac) actually make spatial measurements. Now, these room EQ systems may, and best of them certainly do, have clever algorithms to get the same result from 9 spatial sweeps vs the 50 samples MMM RTA. As we don't have details of their algorithms we cannot use their 9 points method to control the results as they have done more "magic" with them than a simple average. But luckilly for us we don't need to do it their way - we can simply make MMM RTA and get very acurate picture of FR. We can also take a number of sweeps at different points, apply vector average and some FDW and we will get very acurate phase response of the correction.
Regarding time corection - it is nice you showed step response as a proof of time-alignement of the drivers and also that no pre-ringing was introduced during phase correction, but the truth is those things simply don't matter much in audible terms, no matter how nice they look on the various time-domain graphs. There is one thing that might, and very often it does, cause an audible issue but you didn't show it, nor todays room EQ systems are addressing it with their time domain correction, and that is frequency response cancellation in LF caused by phase mismatch between speakers. Unfortunately I have yet to see modern room EQ systems that tries to correlate the phase response of left and right speaker to avoid cancellation.