And yet such a feature would more or less solve the problem for a large majority of users. What you describe as terrible seems like a pretty good feature to me. If a URL is visually indistinguishable from amazon.com yet differs at the byte level its probably not legit.

If I were implementing it I would render the domain text and then check how significantly pixels differed from its nearest "known" domain. We used to do this with render tests where there was a bit of noise.

Don't let perfect be the enemy of good.