None of these demo binaries will actually run on my OS, so maybe the extra 768 bytes are a fair tradeoff for abstracting away the platform.
Interestingly this runtime happens to be way smaller (2MB) than Firefox, Chrome or even Safari.
I understand your irony and do agree that using standards of today is expected.
This doesn't remove any validity to the OP point that what we use today (apart from WebGL or other hardware accelerated code) is several orders of magnitude behind what was already there 10 to 15 years ago.
I do hope though that we'll see more and more widespread/standard techniques to grab back a bit of speed there without only counting on Moore's law :)