Many of us who wrote articles for the Bell System Technical Journal would disagree. The BSTJ publishers could transform something that made sense when viewed as troff output into unintelligible gibberish. You cannot split a UNIX command line into multiple lines just because it "looks better". Sometimes format really matters.
I think that is true for any scheme -- professionals and editors need to work together. That's what Jon was suggesting. When they don't have shared vocabulary/goals - bad things can happen. FWIW: I can not speak for him directly as I never had this conversation with him (Win might have), but from what I knew/know of Brian Ried I think he might agree with what I'm suggesting. IMO, there will always be cases like the one that you described. This is not particular to any document compiler system. The question is how to bring the two sides together and who has the high order bit?
My complaint with Word and the like, is that the 'control' is hidden. It's $%^& magic -- why is it indenting here? Hey I did not tell it to make it go italics ...
Like Jon and Larry, I'm a big roff fan and still use it. But to give Brian his due, his style sheets were in ASCII and what was happening on the page was fairly easy to deduce. That said, I never used Scribe for anything large (like a book), which I can say I have done with troff. In the late 1970s, I did use Scribe for some papers and found it quite easy to use. Since that time, as a co-author I've also tried the same with LaTex and/or Word and found both difficult. When I have the lead and if I can, I'll use troff -ms with a few extra Masscomp macros (that ORA used to pass on too -- the Steve Talbot extensions for lists in particular).
So from my professional experience, it has been mostly with troff, my use of Scribe was short lived.
I'm pretty sure tht Keller tried to make creating books easier in FrameMaker, as that was one of his target users. But again I only played with it, never really had to rely on it for anything.