I said audio. I provided this as a counterexample to the stated thesis of your post. There exist things that can be done over a network such that latency is not an issue. I am obviously not pulling data over a cross-continental link.
FWIW, the protocols I write at work can do a full data pull - a couple thousand elements and growing - in under a half second end to end. I don't know of any HTML/Web based protocols that can even get close to that over localhost.
So yeah - we know the Web is an utter pig. My point is that it probably doesn't have to be.