Is there empirical evidence that "capitalized" numbers are more difficult to read? I'm genuinely interested, but I do feel like a lot of typography is based on aesthetic assessments of individual glyphs, sentences, or blocks of text as opposed to studies of legibility.
"Tinker compared old-style and modern (lining) figures and found a non-significant advantage for old-style figures in terms of reading speed and error rate. (Oddly, modern lining figures were easier to read at a distance – on a billboard, for example. In other tests, capital letters were also found to be more legible than lowercase when set large and at a distance, so these findings do seem to agree."
From: "Taking it in – what makes type easy to read and why" [1] by Kathleen Tinkel, citing "Legibility of Print" by Miles Tinker, Iowa State University Press, 1963.
When you have a paragraph of regular words, set in lower case, a “capitalized” number is very visually heavy and distracting, unbalancing the flow of the text.
At some point in the relatively recent past uniform-height numbers started to become the norm, I suspect largely because of typewriters and then early computers. (It hasn’t been until relatively recently (e.g. with OpenType) that software could easily set numerals in both the lowercase and uppercase versions. In the 90s, most high-quality commercial typefaces sold separate “pro” fonts which had small caps, old-style numerals, extra ligatures, etc., and the typesetter had to explicitly select the relevant characters and change the font to get those features, instead of just setting a feature flag for a whole text block.)