For me the most important advantage of blind review is not that 100% of papers are effectively blind, but that if you're a noname author you have the right to be blind and not having your paper looked down on just for that reason. That alone justifies double-blind review.
Regarding papers that cite proprietary datasets that no one can access, in fields like AI where it is perfectly possible to release datasets (if there's a specific reason it's a different issue), as far as I'm concerned they should be outright rejected due to lack of reproducibility and inability of the reviewers to check the correctness of the claims. Although I know this is a minority viewpoint and it won't happen.
Regarding papers that cite proprietary datasets that no one can access, in fields like AI where it is perfectly possible to release datasets (if there's a specific reason it's a different issue), as far as I'm concerned they should be outright rejected due to lack of reproducibility and inability of the reviewers to check the correctness of the claims. Although I know this is a minority viewpoint and it won't happen.