You raise a good point in that we don't read letter-by-letter anyway. So, this variant, which leverages recognition of whole words at a time, would have the line tracking advantage of Boustrophedon without the problem of duplicating letters for forward and reverse directions:
THIS EXAMPLE OF BOUSTROPHEDON TEXT WAS
WIKIPEDIA THE FOR SPECIFICALLY WRITTEN
ARTICLE ON THIS OX TURNING METHOD OF
ANCIENT IN TEXT WITH WALL A COVERING
GREECE AND ELSEWHERE
Re:Boustrophedon
by singingboyo on Tuesday October 24, 2017 @05:06PM
(#15545254)
While we don't read letter by letter, this does still require some left-right scanning for longer words. In texts (or languages - German?) with many long words, this would be an issue.
The initial method seems to do the best job of not triggering the part of my brain that wants to read normally - which is a benefit, as I can then use a specific reading method for boustrophedon text, rather than fighting against my initial reactions.
Problem here is that if you enter mid-page it's not necessarily obvious which direction to read.
Here's the example image (with mirrored letters): https://en.wikipedia.org/wiki/Boustrophedon#/media/File:Bous...
And here's a non-mirrored version: