The whole point of LaTeX and TeX is not to worry about the exact format, but to worry about the semantics of the content instead, and let the layout engine decide for you. Like all automation, this generally works very well, except when it doesn't. When it doesn't, the solution is to give it a little manual shove in the right direction, not to rewrite everything in the markup language flavor of the week.
Incidentially, semantics, not presention, was originally the point behind HTML, but it got warped and twisted over time into a kinda-sort presentation-oriented language.
restructuredtext, on the other hand, was originally developed to create documentation for Python programs. It might be good for that purpose (wouldn't know; haven't used it.) But it's certainly not good for typesetting mathematics, research papers, academic quotations and so forth. Hence the large amount of wheel reinvention going on here. It's a little bit like writing your research paper using JavaDoc comments. Sphincter indeed.
I used it a little bit back when I still wrote Python. It seemed a lot like HTML, but much terser, and whitespace-sensitive. ReST could do the same kind of stuff as HTML: build bullet lists, create tables, italic text, and so forth. However, since I had already learned HTML, learning another language that did the same thing felt like a waste of time. I fully believe that Restructured Text is a better markup language than HTML in some ways; however, I simply don't care because the differences are minor, and HTML is so much more powerful.
On the other hand, TeX was developed by Donald Knuth, a guy who spent his entire life doing research and writing papers about it. It has excellent math support, and is a true semantic language. I've written a few papers in TeX and been very happy with it.
Anyway, if RestructuredText were good at typesetting research papers, there would be no need for this project, would there?
You're confusing the language with the build system. LaTeX wouldn't be very useful without the awesome build system. This project is a build system for producing both high-quality HTML and high-quality PDF (through latex) with a single, high-level ReST markup language. It also uses latex formulas for math.
Incidentially, semantics, not presention, was originally the point behind HTML, but it got warped and twisted over time into a kinda-sort presentation-oriented language.
restructuredtext, on the other hand, was originally developed to create documentation for Python programs. It might be good for that purpose (wouldn't know; haven't used it.) But it's certainly not good for typesetting mathematics, research papers, academic quotations and so forth. Hence the large amount of wheel reinvention going on here. It's a little bit like writing your research paper using JavaDoc comments. Sphincter indeed.