I would love to see a study about people that follow C++ best practices. Put a bunch of C++ devs and tell them to write some programs. Then see how many of those programs would be valid according to rust's borrow checker.
Whatever % of people that "fail" this test, is much higher than the 0% of people that would do so using rusts' compiler.
Of course, programs that don't pass the borrow checker can be totally memory safe, but that would need to be analyzed on a case by case basis.