The thing is, you haven’t really shown that these sites aren’t successfully optimizing for conversions. Couldn’t it very well be the case that UI which annoys some high-intent users by interrupting them or adding steps to the checkout process also increases overall conversions?
True, A could be “annoy users” and B could be “don’t annoy users”, and A could perform better overall, but in this framework you might be missing C which is “annoy users except those already deep in the funnel”.
My point is more that there can be two groups of users with mutually exclusive desires, and it can be practical to choose to satisfy one group over the other based on your particular goals. There's not always some monotonically increasing function where you can over time satisfy a higher and higher portion of your users.
