[ghero]

I don't think this method is rugged enough, given that it sacrifices the originality of the content, I can see it to be broken systematically in a following process:

1. purchase 2 copies of a book, book A, book B, by 2 different usernames, so by the SiDiM standard, A, B should be slightly different.
2. Extract content from A, and B.
3. Perform a sequence alignment (or sth similar) on the contents of A and B, this should tell where in A and B, the contents differ.
4. Randomly select 50% of the positions where they are different, and swap the differences there, producing A', and B'.
5. Select either A', and B', and both copies should be no longer tracable.
And it is done.

Also there is a cheaper way of getting around this by approximation, it is to get 2 diff copies of a book, and select each page randomly from the 2 copies.

And if this is not cheap enough, just buy only one copy, and add noise into it. Like normalizing or randomly switching grammar, spacing styles. Even if it only reduces the tracing precision by a neigbourhood of 33%, I think this will result the tracing method to be untrusted in the legal court.

Maybe you may say, this might caught unsuspected illegal distributor, but many other (if not any other) DRM method do this just well...

In the end, I don't think the improvement worth the sacrifice of the orginal content. I just don't like a method that is doing this kind of damage.