I take it you are referring to
this "experiment"? Rowbotham has mistakenly assumed that the datum is being measured as a chord. I think you would find that, if you asked a surveyor, the datum for a large work such as this would be relative to sea level; an arc on a round earth. This would still give you a shorter distance covered by the datum line (if the line is entirely below the level of the work being conducted), but not as short as a chord, and quite possibly not short enough to be significant, especially if the distance to be covered has actually been measured over the surface anyway.
A little maths to show what I mean:
For a 1,000km railway with a datum 10m below (and parallel to) it's path, the rails cover 1/40 the earth's circumference, or pi / 20 radians (this is going to be important).
The datum covers the same arc in radians, but with a smaller radius: 6,366.188km instead of 6,366.198km.
This leaves us with the datum being 999.998km long, a difference of only 2m in 1,000km.
I don't think they're going to worry too much about that! Thermal expansion would make a much bigger difference over that distance anyway.
Going on Rowbotham's mistaken belief that the datum is measured as a chord, the datum would be 995.893km. Yes, that's a significant difference, but it's not how it's done.