Via Wes Felter, I came across a fun puzzle language by my old friend C Guy Yarvin. I think of Nock as being in a family of languages with certain things in common: tiny number of primitives, definition can fit on a t-shirt, not quite practical, but powerful and expressive enough that you can imagine writing real programs in it. If this sort of thing appeals to you, I recommend checking it out. I took up his challenge to write an interpreter and sample program, and the core Python interpreter is 38 lines of Python.
All this reminded me of my own Io, which I designed, oh, about 22 years ago. Io is basically a language in which the only primitive is continuations. Because it encompasses the power of simple lambda calculus, you can implement arithmetic and data structures using Church numerals. Probably the best introduction to Io is the chapter out of Raphael Finkel's book, and there's some more good description in notes by Kragen Sitaker.
I only wrote up half of my ideas in the SIGPLAN Notices article (PDF). The other half was primitives for parallelism. In addition to the pure language, "par k1 k2" fires off continuations k1 and k2 in parallel. Then "join k" invokes k with two new continuations, call them "alpha" and "beta". If one thread calls "alpha f" and another "beta x", then the two threads join at that point and invoke "f x". With this primitive, you can easily implement the full range of concurrency operations, including cells with state, input and output (hence the name of the language), etc.
I used to think it would be a good idea to build real systems in Io, but now I guess I've grown up a little. For one, while I think it's important for any serious student of languages to understand CPS, I'm no longer a huge fan of actually using it. Plus, it's much harder to implement things like Io efficiently than I originally thought.
But these languages sure are fun to play with!