It's been done before. The underlying issue with why this isn't a goal is the underlying design philosophy. The general belief seems to be that reader macros are more trouble than they're worth, and that tagged literals provide a controlled way of opting in to messing with the reader, rather than wholesale replacement or introducing a large, non-obvious surface area to mess with language semantics.
Given the fact we have tools.analyzer, tools.reader, and instaparse, implementing new reader syntax on top of the language is perhaps ameliorated by libraries.
Personally, I've dipped pretty hard into the CL read table abuse (to bootstrap clojure for instance), and I'm less enamored with reader macros than I was when I first learned about them. Seems to be a non-issue with Clojure, and perhaps a novelty in CL except in instances where you're actually doing language research and want to hijack the reader to make it easier to bootstrap stuff. In the absolute sense, it's a flex to demonstrate the "power" of languages like CL that allow messing with the reader over Clojure, but in practice, it's really underwhelming IMO.