hans aberg writes

    There appears to be two variations, one based on the original TeX, and one
    with TeX having some kind of extensions.

    As for the second approach, it seems me that the internal representation
    should be 32-bit Unicode. As TeX does not seem well equipped handling the
    encoding issues, one should then hook up a preprocessor providing the
    suitable translations. Thus
        whatever encoding -> preprocessor -> UTeX
    This easy-to-write preprocessor can combine combining characters to single
    Unicode characters, if possible, or otherwise write them on a form that
    UTeX easily can handle, say by switching from postfix to prefix notation,
    or whatever. With further tweaking of the TeX engine it could even combine
    TeX combinations such as "--", "---" into single Unicode characters.

while this would obviously work for text in natural languages,
unicode will never contain all the possible "embellished" letters
and symbols used in math.  (and this may include instances with two
or even more diacritics on a single letter or symbol.)  this set,
while not infinite, is much too large to want to address even using
the unicode private area.  but for latex (or any successor) to be
useful for the particular content for which tex was first developed,
this has to be taken into account.
                                                        -- bb