I think the key difference is you get to that incrementally. You start by "making the monkey dance" on an otherwise-static page, and you can work your way up to rendering everything on a fully-programatically-controlled canvas, a little bit at a time. Forcing you to do everything yourself from the start as TCL (remember tklets?) and Java did makes the barrier to entry much higher.

It's fairly rare to have web UI's built with canvas and WebGL - usually because they need to do something that is impractical using the DOM. They'll still use the DOM for the more conventional parts of the UI.