What do you mean with variability of observers?
I believe that it is possible, even feasible, to have an object (e.g. a painting), and two people A and B, and two prints with the following properties:
Both A and B agree that the two prints are different. Each thinks one of the prints is a better match to the painting than the other.
But they disagree about which print it is.
If indeed this is the case, then fining your process down to give reproducibility and color accuracy substantially greater than the differences between these two prints (however you choose to measure it) strikes me as futile. I have no real sense of what the scales involved here are, though, hence my original question. Is it Crazy Hard to get this close? Or is it only Moderately Hard and beyond that it's about pleasing oneself? Or what?
Something I rarely see mention of is comparing the color in the print to the actual object photographed. This is impossible with, say, a landscape, as the light will never be the same. Color Management is often reduced to the problem of making prints that look as close as possible to the picture on the screen, without reference to the original scene or original object. It is reduced to managing mappings between color spaces, without reference to the physical full spectrum colors that are, ultimately, the source of it all.
Reproducing Art is, on the other hand, ALL ABOUT the latter.