I don't think people who really understand referential transparency would claim that it's impossible to give C a referentially transparent semantics. It's pretty easy. (This essentially feels like looking at the meaning of the language from "the inside" versus "the outside" if that makes sense.)

There's a difference between being able to imagine such a semantics and having that semantics be in common and widespread use or having it be useful to explain functions of the language or common techniques.

The nice thing about, e.g., Haskell is that these referentially transparent semantic models are really just shoved into your face. Purity makes it hard to ignore referential transparency. Monads make it incredibly clear when you have regions of code which aren't referentially transparent.

So, much like the whole contentious "vacuous type system" arguments from a day ago---these analyses exist naturally for practically any language you can think of but their value varies a lot depending on whether or not they are natural.