"Under an author pays model, we estimate the actual cost per paper published would be in the region of £10-£30,000 depending on the impact of lost advertising."

"There are many answers because there are many journals for many disciplines, and the impact will be different depending upon which discipline or which journal you are talking about. In our letter to you, speaking on behalf of Nature Publishing Group, in the case of Nature itself, the British international journal, in order to replace our revenues you would have to charge the author somewhere between £10,000 and £30,000 because the costs of editorial design and support are so high. The reason for the big disparity is how much advertising"

Although subsequent media reports failed to mention it, the quotes above make clear that this figure is only claimed to apply to Nature - an extremely special case among the tens of thousands of life science journals. Elsevier's evidence confirmed that, even with the inefficiencies of publishers' current systems, the cost per article for a typical journal is far lower:

" The cost to publish an article [...] ranges from between $3,000 to $10,000 per article [...] I would agree with those numbers."

" For Blackwell? [...] it worked out at £1,250 per article. That was the cost of the total system."

"Very crudely, £30 million of sales: we get income of £30 million and we publish 1,000 papers a year. That is your [£30,000]."

1. A significant fraction of Nature's £30m revenue is spent to commission and produce the non-research-article content of the journal (e.g. News & Views articles, book reviews, commentaries, editorials etc.) This non-research content would continue to drive healthy print and online subscription revenue, even if the research articles were made freely accessible online. Since the non-research content (the front-matter) is far more widely read than the research articles themselves, it is far from clear whether making the research articles Open Access would have any negative impact on subscription revenue. In fact, the opposite can be argued.
   
   
2. For the same reason, there is no reason to believe that Nature's impressive advertising revenue would suffer dramatically as a result of Open Access, yet they are assumed to fall to zero in Nature's calculation.
   
   
3. Part of the argument used to justify the high cost per published article is that Nature rejects more than 90% of papers submitted, and so has to review more than 10 papers for every one it publishes, and has to bear the entire cost of this.

"[Nature] publishes fewer than 10% of the research articles submitted. Economics dictates that high quality journals like Nature have a high unit cost per paper published, because for every article published more than ten have been reviewed and de-selected."

Letter to Inquiry, January 13th 2004, Richard Charkin (CEO, Macmillan)
This would indeed be expensive, and it is true that the repeated peer-reviewing of rejected papers as they trickle down the journal pyramid is one of the worst inefficiencies of the present system. In fact, however, Nature is not that profligate and had already taken steps to address this issue:

"If a paper cannot be accepted by Nature, the authors are welcome to resubmit to Nature Cell Biology. Nature will then release referees' comments to the editors of Nature Cell Biology with the permission of the authors, allowing a rapid editorial decision. In cases where the work was felt to be of high quality, papers can sometimes be accepted without further review"

From the HoC website
Thus, if a paper is scientifically sound, but is not exceptional or fashionable enough to appear in Nature, it may well be submitted and accepted into one of the next tier of journals in the Nature stable (Nature Cell Biology, Nature Medicine, Nature Genetics etc.) without requiring significant additional editorial work or costs. This is a very sensible system, and is one that is already in use at BioMed Central. If an article is rejected for publication in BioMed Central's top-tier journal, Journal of Biology, but is judged by the reviewers and editors to be scientifically sound, the authors may be offered publication in one of our more specialist journals. Public Library of Science plans to operate a similar mechanism as it launches new journals.

This trickle-down approach benefits authors by avoiding the delays caused by repeated rounds of peer-review, and benefits science as a whole by reducing the cost of the publication process while maintaining quality.