You could make things much more readable by just using decent formatting and descriptive variable names. I know, it's decorated C code, and academic C code at that, but damn, some of that stuff is a mess to try to read.
Maybe I've been in C#/Java/Python/JS land for too long, but javadoc/xmldoc/docstring documentation is a lot more readable. If you really want to, you can dump in markup, pictures, non-ASCII characters, etc into the html-based documentation.
The more I look at it, this stuff must be a monstrosity to try to actually write. I can't see it being at all attractive, unless you were doing well-defined, algorithmic code. Or you could write it in normal C and package it as a library that people might actually use.
It's hard to miss that it's on Knuth's website. Just because it's published by one of the gods of computer science doesn't mean that it is code in a style that anyone should be encouraged to emulate.