A/B testing is not the correct way to test what is objectively better. As we can read from here, people often have personal preferences to what they are used to and think are best. Unfortunately what people think is best is not necessarily same as what is objectively best for them.

If you put people into usability lab and test their reading speed and reading understanding with different text sizes etc. you get more objective results (that mostly agree with the OP as far as I can remember).

Hi; I love your writing! I can't speak for the OP (though I doubt they A/B tested), but most typographic conventions come from the days of typesetting and the subsequent studies of typographers and graphic designers in the 20th century.

If you haven't already, I highly recommend that you read Josef Mueller-Brockmann's Grid Systems in Graphic Design/Raster Systeme fuer die Visuele Gestaltung; the content is really right up your alley. Mueller-Brockmann gives the "rules" of typography, then gives examples of what happens when one deviates from those rules.

You'll also see from his book that the main problem with web design today is the lack of use of columnar text layouts. Blowing the text size up on web pages is just a hack to fill space; if more responsive layouts used multiple columns, that kind of trickery would be unnecessary. (Edit for clarity: I'm talking about multiple columns for a single body of text, much like you'd see in a newspaper or magazine.)

I can't speak for the OP, but there have been studies of appropriate line lengths for reading text on computer displays. See for example https://web.archive.org/web/20020725135307/http://psychology... (found via WardsWiki: http://c2.com/cgi/wiki?TenWordLine).

As you might expect, people who can read better can generally read longer lines, limited mainly by the number of words apprehended in a single "fix" or by the reader's visual acuity.

The article I link to references research on both physical line length and characters per line, but the methodology described in the article itself attempts to fix physical size of a character and distance of the eye from the screen.