For homogeneous text like in a novel, snapping to a grid may be
 considered quality typography. For material with lots of odd-sized
 objects such as equations or smaller-typesize quotes or section heads,
 snapping to a grid can easily lead to vertical spacing that looks
 worse than the usual TeX approach.

I don't agree. It is very easy to have section heading withs 1.5
\baselineskip above, and .5 below, e.g. If you're using a multicolumn
layout, facing lines that disagree are horrible _and_ disturbing.

What is very bad is TeX's behaviour of linespacing the lines with
overhanging material (like integrals, badly formatted inline
fractions...). I believe to be the responsability of the editors to
"flatten" inline formulas, and display others.

 But if you assume that all lines are normal height and all vertical
 spaces are integral multiples of baselineskip, a grid-based system is
 easy. However ... if you also want automatic widow/orphan suppression
 *and* flush-bottom columns then you are talking about mutually
 conflicting constraints which are hard for any system, not just
 LaTeX. Most publishers just have human beings fixing the more
 difficult page breaks "by hand".

Yes, of course, I know i asked too much ;-)
However, TeX (or any good typesetting program) _should_ be able to
reconsider a paragraph whose line breaks occur at page breaks, e.g.
rather than simply pushing the line to the next page! same for widows,
clubs, etc.

Sometimes, when you take control over every line break in a paragraph,
you see that tex could have done it like you, according to its
penalties... That's why I'm seeking "better but practical"

Also, It seems so obvious that double pages are the unit of book
composition that it looks weird to me that TeX simply ignores
that. Imho, this should be handled at TeX's level rather than
LaTeX's. But anyway, yes, asking the shipout to do much more
computation may be a temporary emulation of a better program?

Thanks for the interesting comments!
Thierry Bouche, Grenoble.