The state allows parts of program to communicate through implicit information channel. One part of program could signal another to increment a variable by deleting a file. Or vice versa.
That's why I think Haskell is the best for libraries, Scala less so and Lisp even lesser, even taking into account macro system, code-as-data, etc.
That said, why not try Clojure? Best of both worlds, and then some.
But I'm still at the point in my career where people hire me rather than the other way around, and all the hand-waving I can do about "superiority of lambda calculus over the state machine" (okay, I'm more clear and practical than that) doesn't convince many.
Making Python or Scala talk to Oracle (for example) is so trivial you can write that bit of your program in no time at all, then get onto the interesting stuff. In Lisp, you have to spend significant time on this trivial stuff. So if you are writing a program on your own PC for your own use, then Lisp is fine. If you are programming anything that is going to run on existing infrastructure or interoperate with other programs, it's very difficult to justify the additional effort of using Lisp.
And I know a bit about this: I wrote Oracle bindings for OCaml! It was fun, sure, but can I justify it when there are already Python bindings? Hmm.
Remember: Lisp is a family, not a language.
The best counterexample to your claim is: http://clojure.org/
I'd argue even CL covers a lot vs your claim (but please don't make it my arguement, focus on Clojure) with QuickLisp: http://www.quicklisp.org/beta/
Actually, "library" is probably the wrong word for reused Lisp code. A library is a public thing, an institution---and all that goes with it: slow to change, must be easy to understand, etc.
Instead of libraries, Lisp hackers have private collections. Like a messy desk, they are often supremely useful once you get the hang of it, but are inscrutable to others. Not exactly the best characteristics to get popular on github.
Certainly Github does not in anyway support your claim: https://github.com/languages/Clojure
But also its not that its great for writing libraries but that its great for writing libraries that extend the language.
Lisp programmers, when using Lisp for what it's best at, don't use libraries. There are no libraries for what they want to do, because what they're doing is obscure and/or hard.
Lisp is best at doing new things that have never been done before, because it's basically a language for writing libraries. As such, it prospers in unpopular fields.*
This makes it hard for lisp evangelists to be taken seriously. If you're to show someone a hard graphics problem written in lisp vs ruby, your reader has to understand graphics programming first. But calculating primes? Anyone can do that. Which example do you think will get more mindshare?
To summarize: Lisp is great at things you don't hear about.
* One reason Viaweb succeeded using Lisp for the web was because the web was still this exciting frontier and no one had really done anything with it yet. I'd go so far as to say that Viaweb wouldn't succeed today---they'd get trashed by the likes of Weebly, which is written in php for crying out loud.