50% is overselling the success rate of login-walls from Google Images by a lot - probably multiple orders of magnitude. And I think you're significantly misrepresenting the A/B test's goal. It's not to find what works best for the most people, it's to find what works best for the company running the test.

If a login wall causes an account creation from 0.5% of people presented with the bait-and-switch, that's not compared to a 99.5% failure rate. It's contrasted with an account creation from 0.1% of people presented with the image they wanted with a "Click here to create an account" prompt in the corner, and it's five times as effective as that alternative.

The same story gets repeated here over and over again regarding the email newsletter sign-up form that steals focus halfway through a blog post. Yes, they increase your traceable readership more than anything else you could put in your blog (again, counting 1% vs 0.1%) . But there is a large percentage of the population who will close your blog permanently if you use them.