As eric256 suggests, Tim's looking from a book perspective. The picture is a lot more complicated; consider that O'Reilly published only a handful of niche Perl books for a couple of years while expanding and updating the available PHP and Python titles.

This is not a research project untouched by the observer. Releasing a new edition of the Camel or the Panther or even the Llama affects the market profoundly. (Is a Camel from 2000 still viable and valuable? Yes, but try convincing a bookseller that in 2003, 2004, 2005, or 2006. I believe you'll have a more difficult time every year, even if you can demonstrate that it outsells even more recent books.)