TMAX100 had granularity of 8, not 18, but I agree that Tech Pan is better  on that measure; my conclusion is that the two are roughly comparable overall, desire advantages to each is various respects.
You're right, I overlooked the figure  at halfway the document.
The fact that most sensors use Bayer CFAs, and therefore sacrifice some resolution to CFA and demosaicing, just strengthens my point: ...
Sorry, I disagree (explained below).
... to compare the "sensor resolution" that many B&W film users found worthwhile to the sensor resolution currently offered by things like the 7,360 x 4,912 pixels (36MP) of the D800E.
And that exactly illustrates why I disagreed above. The 35mm film scan @5400PPI gives 7654x5102 pixels and the D800/D800E gives 7360x4912 pixels. However, one cannot compare
the 'sensor' resolution in isolation. One delivers a partial product that will degrade further upon scan/print, and the other is already output 'ready' and will not degrade at all.
Demosaicing goes in the balance on one side just as scanning or the effects of printing with an enlarger go on the other side.
The degradation is very much incomparable in magnitude.
This is almost verbatim from The Theory of Constraints. You maximize your improvement by throwing all of your resources at improving the weakest link.
Absolutely correct, it always helps to improve the weakest link. However, in the case of cascading MTF responses, a combination of 50% and 40% response at e.g. MTF50 gives a combined MTF of 0.5 x 0.4 = 20% . Improving the worst of the two (40%) to 50% will boost the combination to 0.5 x 0.5 =25% MTF, and boosting the best (50%) to e.g. 62.5% gives 0.625 x 0.4 = 25%, IOW the same result.
So in the case of cascading MTFs it can also pay to improve the second weakest link instead (although it may be cheaper to improve the weakest, or not).