This is a very good point - Lua seems to have improved quite a bit since (say) 4.0 - it's small enough that maintaining an in-house fork of on old version is an option, so the language designers can make major changes when it's a clear improvement.
As library maturity goes, there are some really good ones out there (such as LPEG and LuaSocket), but for some things the documentation is a bit thin. The situation is getting better, though.
This is exactly the problem, imo. Lua is a niche language because its benefits only apply to a certain niche, i.e., scripting glue on top of an extant system, because that's where the time investment makes sense.
Is Lua so much better that using it is worth losing the convenience of something like Rails or Django? Those things come all bundled up nicely with everything you need for a website. Is it worth the investment to go around and scrape all of these pieces together in Lua? Is Lua going to provide a benefit commensurate with the time you invest in piecing all of that together yourself, including the real possibility that you'll have to write much of the functionality yourself? Or is it simply cheaper and smarter to use Python or Ruby where the frameworks are already composed and widely tested and deployed from the get-go? There's your answer.
To the extent that one wants to increase adoption of a language in a certain niche (say, web programming), you need to at least attempt to make it basically competitive. However cool and fast Lua may be, for most people, it's not worth the investment to start from scratch on a framework when Python has Pylons, Django, Pinax, and Ruby has Rails, etc.
lua is not a one stop shop and will never be, which probably why it flys under the radar so much - you cant stamp a powered by lua logo on it.
Also, the idea reminds me of this: http://c2.com/cgi/wiki/wiki?AlternateHardAndSoftLayers
And there is a c/++ library for almost everything.
QScript, ChaiScript, and projects like boost.python and libperl++, play much better with your C++.
I've considered a number of these for my current app, especially QtScript, Lua and Squirrel, but eventually I settled on embedding a JVM, since it gives me access to a bunch of libraries and lets me use any JVM language (I'm personally using Clojure and Yeti, but users of my app could in theory use other languages).
The problem is that it isn't a good "starting place" for an app. Instead you write something in C/C++ and say "oh, I wish I had a dynamic language for this part" and then you add Lua on top. And this is part of why Lua is such a compelling language now - by being extension-focused it's been able to iterate the entire language many times over and make basic semantic changes without worrying about backwards-compatibility.
But if I start writing an app and want something Lua-like....it ends up being easier to start in Python. There are some good projects out there to address this but it's a question of maturity - as of right now Lua still doesn't really satisfy as a "batteries-included" platform.