The premise behind Racket, in broad strokes, is to provide a platform and programming language for making new programming languages. The most important identifier/keyword in Racket is '#lang' (specified at the top of a file to tell Racket which language the file is in). It's a whole ecosystem in which a multitude of languages exist and the only thing they really need to have in common is that on some level they speak Racket.
Personally, I think the above paragraph explains how exciting Racket as a language and platform is, but it doesn't sound that interesting on the surface. It does allow people to create things like this[0] and this[1], though, which displays real practical use; shaping the way you solve the problem to fit the actual problem.
[0] http://www.youtube.com/watch?v=oSmqbnhHp1c - Naughty Dog's scripting language [1] http://pollenpub.com/ - A web book publishing language
Nearly every imperative language is a walk in the park after you know C++, and you can still write C++ for performance-intensive code (if necessary with expression templates et al.).
[ed: and perhaps PostScript/Forth...]
Overall, I think the more languages that exist, the better. Some people think we should be moving towards more domain specific languages.
There is a balance between using the best tool for the job, and getting a large potential developer base.
Are there more now? I remember a zillion niche pet languages from forever ago.
Now a days I see less, probably because it's not in-style to write an in house scripting language for every single new project, they tend to just use a language that is nice off the shelf.
But, the Arc compiler is written in Racket[1], and internally outputs Racket code[2]. "mzscheme"[3] is an early version of PLT Scheme, aka Racket.
[1] https://github.com/arclanguage/anarki/blob/master/ac.scm