The vagaries inherent in selecting best-sellers do not really stem from problems of accuracy.
At present, there are only two entities that track overall sales of a book: the publisher, which tracks books shipped and/or returned, and Bookscan, which kind of sort of tracks books sold at the register.
The publisher's numbers are as accurate as reasonably possible for the very simple reason that they have to pay the author based on this number and are subject to audits at the author's request. However this does not track copies sold to readers--just to the indies, the chains, and other retailers. This number is occasionally made available to industry press (for example, Publisher's Weekly).
Bookscan numbers track copies sold to readers, however depending on the genre Bookscan may report 90% of sales or it may catch 50% of sales, because not all booksellers report to Bookscan. Bookscan subscriptions are not cheap to get (publishers and some agents are their primary customers) but Amazon recently made authors' personal numbers available via author accounts.
The best-seller lists rely on a combination of publisher input, Bookscan numbers, non-bookscan numbers, and their own statistical projections. My wife, a New York Times bestselling children's author, has spent some time examining the lists and the numbers with some of her author friends. Most of the time, Bookscan numbers line up more or less with the rankings on the list, particularly at the top. But (especially toward the bottom half of the list) there are recurring and sometimes wild variations. And some books are not "listed" because the publisher apparently does not submit them for consideration in a given week; Lois Lowry's "The Giver," for example, puts up strong enough numbers every single week that it would likely bounce on and off the Children's Hardcover list several times each year. Because the list is a powerful marketing tool, however, and "The Giver" presumably is in no need of marketing, this does not occur. Furthermore, the NYT has shown that it frowns on books clogging the list for too long (the Children's lists were made in direct response to Harry Potter's dominance, for example).
With specific regard to e-books, we're a little baffled as to why the NYT would create an eBook list and a "combined" list at once. I don't know if eBook sales previously counted at all toward a book's listing status; I do think they should! I can see why a separate eBook list might be of interest but I'm not sure THREE lists is warranted. As a general rule, expect to see the top slots of all three lists basically repeat themselves. But don't expect "improved accuracy." While accuracy is definitely among the lists' aspirations, the ability to track eBook sales only slightly improves the information already available to the Times.