The failing is when HTML/JS needs to do 'heavy lifting' that native code does better (such as large memory based tasks, 3d rendering for now, animations, heavy math, or hooking into hardware).
For example, I've never played a JS/Canvas game for the sake of it being a good game. Has anyone? I see a lot of frameworks, and a lot of demos to advocate for the platform, but no good games. Why is this? At this point, the novelty of the platform is no longer an excuse. It has been a forum darling for several years now.
First, canvas really needs to be hardware accelerated (someone already does this, Microsoft?). Having a few objects move around should barely take any CPU but with Canvas today it can be anywhere from 10%-20%.
Second, their needs to be game engines/abstractions on top of Canvas. Most developers don't want to have to write there own render/clear loop with all the utility functions to do movement, render shapes, texture elements, and keep object state. I know their are a few out there now but they are no where near the quality needed. This second step will not happen until my first point (HW acceleration) is in place.