To my knowledge, it cannot be done in a way that addresses the privacy concern: that a court could require the storage operators to answer such questions about their customers as described in the article.

If there's a way around that, I am your eager student.

By the way, some people have suggested using reference counts as a solution. You then know, for example two or more users share duplicate data, but not which two users. My feeling is that, at some time, you did know which users those were (when the counts were incremented.) I like to stay within the realm of "we are not capable of knowing" instead of simply, "as it happens, we don't know anymore."