nacl

CurveZMQ (basically, DJB's CurveCP over TCP) is probably a better choice if you want a NaCl-based secure channel.

Code compiled for NaCL is in a word-aligned subset of machine code that can be quickly verified before execution and can only generate interrupts to the NaCL host (no syscalls).

With the advent of things like NaCl, Emscripten, and WebAssembly these things should not be necessary anymore.

Very different from {P}NaCl, which was very successfully killed by everyone except Chrome ignoring it.

A dissolved mole of NaCl produces one mole of Na+ and one mole of Cl-.

The client would be a platform independent bytecode of some sort, like the NaCl.

Yukihiro Matsumoto has only worked for NaCl according to LinkedIn and Wikipedia.

As you correctly state, NaCl requires each nonce/key pair to be unique.

I'll keeping using ChaCha20 when I need to (it's part of nacl/libsodium if you don't know where to get it from).

Is the main advantage over TLS in its use of a limited set of cryptographic primitives (as provided by NaCl), at the expense of flexibility?

NaCl's ChaCha20-Poly1305 is even better, because it's fast and constant-time.

But the NaCl key is randomly generated for each connection.

I've not looked into NaCl much but maybe someone can answer me this: running "real" code in a browser but sandboxed for security?

NaCl isn't portable (Mozilla is working on Web Assembly) Dart excerpt taken from Dart FAQ; question: "Will the Dart VM get into Chrome?" response: "No.