I agree with you that it isn't adding up. The question is why. And I agree with you that if Canson doesn't have Abdo's problem, it would be good to know why. Someone should ask whoever does this for Canson how they make their profiles. I don't know who that would be otherwise I would do so myself.
I think the nub of this problem of differing results stems from two places:
(1) Windows - if I understand correctly - does still allow the user to disable ICM so that the OS does not tinker with file values on their way through the computer to the printer, but Apple eliminated the option to turn Colorsync off in OSX, as Ernst rightly pointed out. This means that printing correct targets on a Mac requires a workaround not needed with ICM. It led to the development of ACPU after the uproar that happened when Adobe turned off the ability to disable color management in the Photoshop print function as a result of Apple's action.
(2) Canon appears to have designed their drivers for the Pro series printers in a manner that interferes with the proper functioning of ACPU on those printers, but Epson has not.
Both of these situations give rise to the observation I derived from advice and experience that it is best to use ACPU for Epson printers and Print Studio Pro for producing targets from Canon Pro printers. These two paths for those two printer brands have served me well in the many profiles I have generated for the printer and paper reviews I produced for this website and for my personal photography.
Possibly underlying the issue Abdo raised is the point Doug raised that there may be some differences of settings that is affecting whether colour management is actually being turned off to the same effect in both operating systems with whatever he is doing to produce the targets using each. Let us recall that a profile is characterizing how the printer renders file values, so if there is REALLY no colour management happening in both operating systems, the same printer printing the same target on the same paper should produce targets that look the same. If they don't, "No Color Management" doesn't mean the same thing between the two OS types, and the reason for that could be a user-controllable setting, or something deeper under the hood we haven't gotten to the bottom of. I realize this post isn't settling the question, but I hope it contributes to further thought on the etiology of the issue and where to look for conclusive answers.
Mark, I think you summarize the conundrum quite well.
For Canson's profiles, which are identical for Windows and Mac, to work the process to create them, as well as use them, simply has to result in the same down to the driver level. Yet, your testing shows some differences on the Mac with ACPU v Canon's plugin. I don't have a Mac and you don't have a Windows but each of us have the tools and background to clarify this. Abdo has both, but is new to this. I can appreciate his confusion.
From my pov, this is quite amazing. I have never encountered an issue with Windows producing any differences with ACPU, I1Profiler's direct printing, or the null-transform Photoshop trick. They all produce the same results properly done. However, each approach has its issues.
ACPU, for unknown reasons, shortens the targets on Windows about 3%. In my testing this creates targets that are just acceptable to ISIS. While it works, if the targets are only 1% smaller than ACPU prints, ISIS fails to read them. I'm uncomfortable operating on the edge.
The Photoshop null-transform trick, which has always worked on my printers with all the various flavors of Photoshop and Windows, has been deprecated by Adobe to the point they raise a stiff warning when doing it. Most likely to keep a common code base with the Mac and deal with the fact it hasn't worked in some time on the Mac side. However, for me it's indispensable when printing targets on large roll paper where positioning multiple target pages across a wide roll and having them printed and sized correctly, just makes that process much simpler.
I1Profiler prints targets directly that are sized correctly though it's limited to 8 bits and is somewhat difficult to use on wide roll paper without wasting paper. Also, when printing targets with fractional values scaled to 0-255, it truncates the fractional values. Yet, when creating profiles it uses those fractional values. This is only a problem when generating and using targets that contain fractional values. Unfortunately, I1Profiler produces profiles using fractional values, but saves the truncated ones inside the generated profile. The workaround is to save the generated targets which also truncates, then reload prior to printing the targets. Then the target prints and profiles are created from the same data. This is fine for profiling smaller sheets but wastes paper on large rolls.
As for the other printer settings such as resolution, paper thickness, type, vacuum, and whatnot, I simply optimize inking (another process that can sometimes be useful but separate from profiling) then save and name the configuration. Then, when printing either targets or using Photoshop to manage color, I load the appropriate named configuration. This approach works great on everything I have with both my Epson 9800 and Canon 9500 Mark II.