I'm guilty here: I described this format as "compression technique" because early feedback indicated that many people assumed this was a new bytecode. However, the main objective is indeed to speed up parsing. Compressing the file is a secondary goal.
> My suspicion is it's going to be marginal and not worth the added complexity for what essential is a compression technique.
In terms of raw file size and according to early benchmarks (which may, of course, be proved wrong as we progress), Binary AST + gzip affords us a compression that is a little bit better than minification + gzip. By opposition to minification, Binary AST does not obfuscate the code.
The real gain is in terms of parsing speed, in which we get considerable speedups. I do not want to advertise detailed numbers yet because people might believe them, and we are so early in the development process that they are bound to change dozens of time.
> This project is a prime example of incorrect optimization. Developers should be focused on loading the correct amount of JavaScript that's needed by their application, not on trying to optimize their fat JavaScript bundles. It's so lazy engineering.
Well, you are comparing optimizing the language vs. optimizing the code written in that language. These two approaches are and always will be complementary.