I find that the best full frame digital is about the same quality as 645 film, maybe very slightly less. That assumes a good tripod and mirror lockup. 645 digital is awesome, but also $16000. Film camera resolution is a function of both lens resolution and film resolution, which depend on and affect contrast. Digital is a bit different, more or less giving you the minimum of lens resolution or pixel resolution. The pixel game can become meaningless as lens resolution becomes a factor. A very good general purpose lens will only resolve 120 lines per mm for a high contrast subject. Most resolve less. For a 24X36 sensor, that's only 2880X4320, or 12.4 megapixels. The combination of film and lens usually only gets you about 80 lines per mm, maybe 95 with the best equipment and common film.
The idea that resolution is limited by lens resolution and that the good general purpose lens will only resolve 120 lines per mm (line pairs, probably) would lead IMO, following this reasoning, to saying that FF digital cameras with more than 12.4 mp don't have a higher resolution than those with 12.4 mp. If the lens doesn't resolve more than 120 lp/mm, and that must be true for digital and film, why would FF digital camera have so many pixel?
(it goes without saying that all this reasoning implies tripod, mirror lock-up etc).
That seems to be contradicting not just what the industry does (that might be due to marketing hype and consumer irrationality, although I don't think a consumer of a 4000 Euro camera behaves irrationally often) and the general experience of photographers (who do use high-mp sensors mainly to be able to crop the image without going below a certain mp threshold, which means it does make a difference) but also the original affirmation.
I don't understand how this affirmation would stand with the other, "
I find that the best full frame digital is about the same quality as 645 film, maybe very slightly less". It is my impression that according to the "120 lines per mm" reasoning, a FF (24x36) digital should not go beyond 12.4 mp (your figure, unless I misunderstand what you say) and I do believe that a scan from a 6x4.5 film image can give you much, much more than that.
You seem to apply your "120 lp/mm reasoning" only to scan from film and not to FF digital.
My affirmation that I do get more than 20mp of real resolution from film stems from daily "pixel peeping" at my scans, for professional reasons (I have to inspect at real pixel size any image that I send to agencies). Sometimes I "blow up" details so that each pixel is a "square" on the monitor, to see how the pixels build the image and if there is some "wasted", redundant pixel, and I can clearly see oblique lines and edges where all pixels "play a role".
If the real resolution were lower than the nominal resolution (given by the pixel dimensions) I would see clusters of pixels with basically the same value where I should instead see an edge, the typical case would be oblique lines where I should see a saw edge. When you inspect images at real pixel size (or, let's say, at 1000% pixel size or so, so that you see any pixel as a solid block) you can clearly see if the real resolution matches the nominal resolution.
More in general, "resolution" is a different concept than resolving line pairs on a mire and this kind of tests do have a role in assessing lens performance but should never be taken as if they were "science". For instance, the general lens has a much better behaviour when focused at infinite than when focused at let's say 1 metre but, as you imagine, resolution patterns are photographed at 1 m (or 2 or what) not at infinite.
When I have time and if anybody is interested (being an hybrid point) I can post some very high magnification crops showing how "no pixel is wasted" in my 4000 ppi scans which leads to a true 20mp resolution from a 135 film. The proof is in the pudding, reasoning about theoretical resolution and lp/mm is no pudding.
As a side note I would like to stress that I scan my images with 16 passes for each location and 2 overall passes with 2 lamp intensities ("multiexposure") so that each pixel is actually the result of 32 passes by the scanner. That "squeezes" 20mp of real useful non-duplicated pixels out of my slides (tripod, mirror lock-up, very good Minolta fixed focal length lenses).
I can see them and all the mathematics in the world will not convince me that my system does not reach that resolution

. Too much lp/mm reasoning leads to the situation where to two Dominicans find themselves when Galileo invites them to look inside the telescope to see the "Medicean planets" and the two Dominicans refuse to look because philosophy is superior to sensorial experience, you know*.
* That's a scene from "Life of Galileo" by Brecht, I don't deem it unlikely something like that really happened.
*** EDIT. Mmh, I looked at a table with lp/mm resolution expressed as ppi resolution.
It says that 80 lp/mm (which you indicate as achievable with good lens, good film and good technique) equals 4096 ppi. So I do accept that my scans have a resolution of 80 lp/mm.
But it follows as a consequence that they do yield > 20mp on 135!