It seems to me that in this way one circumvents possible user-introduced bugs, in this sense: you substitute a TeXmacs construct for another TeXmacs construct, so the converter is not affected.
Maybe (but I do not know) a “direct order”, i.e. "Convert TeXmacs tag x into LaTeX command x'" is more delicate.
The real reason for requiring LaTeX submissions in academia?
Maybe this can already be done using a combination of the tmhtml-my-tag mechanism and the specific tag.
By the way my experience is that for Scheme you “almost need” an editor that helps you both to balance the parenthesis (= does it for you) and keep the code indented (so to allow an easy overview of the structure). I did not try enough the Racket editor to see how it works from this point of view; I am using emacs with paredit, which lets you edit s-expressions structurally. For example with paredit one can “slurp” an item inside a parenthesis or “barf” it outside; I use a part of its functionalities only but this helps me to write things at least twice faster I think.