Got it. Seems he's confusing color resolution with luminescence resolution. Luminescence is far more important to the human eye so his calculation, while kinda sorta technically correct, is not *operationally* correct. His conclusions are awfully misleading. Grr.
That's not how I read it - he arrives at the 87MP figure based on the rated resolution of Velvia 50 in lines per millimeter, and resolving lines is primarily a function of luminance. He then goes on to bump the number up to 175MP once debayering is accounted for - but again, that really only accounts for luminance being half full-res in a bayer pattern. I don't think he's all that far off with his calculations, at least when we're talking about theoretical resolution limits.
But he's also talking about still imaging with particularly slow, sharp 35mm film. Doing a similar type of calculation for typical motion picture film we might start with something more like 130 lines/mm for the film, so we need 260 pixels/mm of vertical resolution, by the s35 full height (18.6mm ) = approximately 4800 vertical pixels to match the resolution of Super35.
Except that Super35 is a 4x3 area, so to compare to widescreen we're looking at only 14mm of vertical image on film for a vertical digital equivalent of 3640 pixels. In a 16x9 image that gives us 6471x3640, or about 23.5 megapixels. For 2.35 it's only 17.8Mp, and traditional Academy format it works out to about 18.3MP. So double any of those to account for a bayer sensor and we're talking 40MP range.
But that's a theoretical maximum for the negative based on a narrow, perfectly controlled situation, and at the upper limits of the MTF curve we're talking about barely resolving contrast between lines - in a test chart situation, where the lines are pure black & white. In the real world, where fine detail generally has less real contrast, some of that detail will visually disappear at a much lower limit. Really fine detail also gets lost in the post process - either traditionally through the interpositive/answer/internegative/release printing process, or via digital scan, and in the color correction process even slight changes to contrast can eliminate that upper limit detail even if it was present on the original negative.
All of which is why 4k is the current common upper limit for film scans - while it certainly may be possible to extract more detail from the original film, it's not likely to make a significantly noticeable difference in the final output.
Of course, to match a 4k film scan takes a higher resolution camera (at least with bayer pattern sensors)... at minimum 8K for a 1:1 match of luminance samples, in reality maybe more depending on the camera's low-pass filter, so we're getting back up into the 40Mp sensor range.
Audiences don't seem to be complaining much about stuff shot on today's lowly 2-4k cameras though, so I'd say a lot of this is academic as far as actually making a good film is concerned. And perceived sharpness/detail probably has more to do with contrast than absolute resolution, so my guess is there's a lot more to be gained by pushing digital cameras to resolve greater dynamic range than film rather than greater resolution.
