The NYT report was based upon the entire universe of published books, for 2003 if I recall correctly (but may have been 2004). That includes everything for which an ISBN was issued—all 195,000 or so items. Considering that what we call "commercial publication" accounted for less than a third of that number, one can reasonably multiply the "per title sales" in the report by three to get a rough estimate. That would imply that the relevant numbers are 3,000 and 15,000 copies respectively—and that is consistent with my own data.
The number is as low as 3,000 not just due to small presses, but to professional and other restricted-audience books, which together form a substantial segment of the market. Consider, for example, law-school casebooks. There are, at present, five casebooks with any real market penetration (that is, they're used anywhere other than the author's own school). The total number of law students taking a basic copyright course—the only place in which the books will be used—is probably around 3,500 to 4,000 in any given academic year. Assuming that every student bought a new book, and that each law-school library in the nation also bought a copy, that means that the total market each year for copyright casebooks would be under 4,300. Even if one book were dominant (at present, that's not the case), at least four would fail to meet that 3,000-copy criterion. And, considering that about 30% of law students show up with a used casebook…