For example, expressions can be strings, so now you have strings inside strings, and the strings may even contain further levels of nested formatting strings. This would make correct tokenization of Rust quite difficult, and out of reach of most syntax highlighting tools.

So the next question is why not allow only some "simple" expressions. It would require teaching users which expressions are allowed and which aren't. The rule that only a single identifier is allowed is both simple to parse, and simple to teach.