> This is recursive and implies a decrease in diversity over time, starting from enormously high and reaching abysmally small. But I am fairly certain that it doesn't hold.
What books are considered part of a canon will always trail behind the total number of books being published, so the total set will either remain stable or increase in size.
However, I think your assumption that it will somehow converge only holds if the criterion of quality are both universal and consistent. Neither has to be true – different cultures have different measures and weights for quality, and within cultures there are disagreement as to what quality is (even in a contemporary setting). Hence my next claim.
> I fail to see what makes old books capable of disagreeing with each other and recent releases incapable. At some point new releases become old and therefore they suddenly attain the ability to disagree with each other?
That's the point – it equivocates the two which allows you to be indifferent between them upto other factors, since you should expect both new and old books to be diverse, rather than the monoculture of thought you are attempting to resist by going completely your own way. Those other factors can include possible selection effects like I have claimed above which act as signals for a book's quality.
I guess the only thing left to add is that you can't read everything, but there's a lot you could read. Given this problem, does one be an early adopter for a book or not?