Performance yes, you can't compete with that one because browsers don't let you control the low-level stuff. But the speed of development and the ease of doing complex layouts with css is hard to match (when you master css enough, of course), specially if you're trying to do UI from the code directly, without IDE helpers and visual builders.

The speed of development is very fast for the first 75%, and comes to a screeching halt for the last 25%. The web stack has no good way to answer basic questions like "does the user have arrow keys" or "what is the current mouse position." That's the point that you start to layer on the "hacks on hacks" that OP references.

Can any system tell you that the user has arrow keys?

> You can create controls in WIN32 or MFC that would simply be impossible in HTML for a reasonable level of performance

Flip side: You can distribute/deploy the host HTML application with reach that will be impossible achieve with a WIN32 or MFC application.

True, but it should be possible to create high performance UI controls inside an HTML5 canvas element.

With canvas, every browser has hardware accelerated immediate mode rendering that could be used to create UI toolkits. Looking around the web, there are some attempts to make this happen, but none of them seem to be hugely popular.

Last time I looked into this (years ago), a big blocker was that nothing inside a canvas element is visible to accessibility APIs, so UI rendered in a canvas would be invisible to people using assistive technologies like screen readers.

Yes, but this doesn't matter for the majority of applications today

Replace "WIN32 or MFC" with "iOS or Android" and the point still stands, especially on mobile devices where processing power is limited and battery life more important.