That paper concludes:
"... the phase distortion audibility results in this thesis research did not seem to be as significant...
The human auditory system was found to be extremely tolerant of even gross phase distortion effects"
Compare that with the Rich/Cochenuer AES paper on The Audibility Of High Order Crossovers (115th AES)
"The second-order LR network received lower
audibility scores than the fourth-order network.
Statistically, all but one test subject could detect it
with noise or music and only two subjects were
unable to detect it with music. This result clearly
points away from the second-order network in the
design of loudspeakers."
(lower audibility scores in this context mean it was more noticable and bothersome to listeners when compared to an ideal source - higher is better)
"It is well established that driver interactions in first order networks create frequency response
disturbances with changes in driver path delays and,
for this reason first-order networks were not included
in our study. These networks also fail to provide
cutoff band attenuation to the extent that drivers may
be pushed towards nonlinear operation by signals
present well outside the crossover region."
"Other than their low cost, phase coherence is the sole reason given by designers for the use of first-order networks."
"Our test results could identify no sensitivity of the
test subjects to phase effects; these effects become
increasingly more apparent as the driver order
increases and our controlled study found the subject's
ability to identify the crossover followed an inverse trend."
"From the auditory experiments presented in this
paper, the authors suggest that speaker designs
should incorporate fourth-order notched crossovers,
with the eighth-order notched crossover made
available in high-performance cases."
(Infinite Slope falls under the fourth order notched with finite zeros category)