I'd rather go with a rewrite that also redesigns the typesetting language. It feels like C++ of typesetting to me, although of course maybe I'm just not familiar enough with the language to see its key concepts clearly.
Maybe the Lout system will pick enough traction one day.
It's a quirky mess, but none of the others, to my knowledge (and I've used most of them, which certainly includes Lout) do nearly as good a job typesetting. Lout does a bunch of things subtly wrong that I can't remember right now but that are very visually distinct, and troff is so far down that road I can recognize troff typeset documents at six feet from a glance.
Lout's equation formatting is especially primitive as I recall, and that is a huge amount of work to get TeX-like quality for. I am unaware of any system that typesets mathematics as nicely as TeX does.
Yeah, I'd like to see someone do a modern rewrite of the whole TeX/LaTeX/XeTeX/XeLaTeX/etc. mess. Something that has modern syntax and support for modern features (such as output to PDF, HTML, even OOXML).
It would be good to see UTF-8 natively supported, and for all the common packages to be absorbed into the core libraries. Right now, you can sometimes find 2 or 3 different packages that do the same thing, often with varying benefits. Take the best package, add the missing features from the other packages, and bundle it all into one distribution. For example, the absolutely horrible table support, where merging cells is a nightmare.
Indeed, once you get set up with LaTeX it's okay, but if I were to go and make a new essay I'd have a hard time remembering wtf packages I used for the last one and what they actually did.
Maybe the Lout system will pick enough traction one day.