I don't have hard data, so I would have just posted this as a comment, but it was too long.
It would be interesting to know who (if anyone) actually pays these per-article prices and under what circumstances. I suspect that no individuals are willing to pay the per-article price out of pocket, so that price isn't actually aimed at them. Most likely the journal's goal is to pick up some per-article sales from institutions; people are often willing to spend their employer's money or their grant money on something that they'd consider exorbitantly expensive if it was their own money.
If so, then we should be comparing per-article prices with institutional subscription prices, not with individual subscription prices. Institutional subscriptions are much more expensive.
Note also that there is a huge glut of highly specialized, low-impact, low-quality journals out there. The driving factor behind the proliferation of these journals is that academics seeking tenure have an overwhelming incentive to publish. Publishers are going to institutions such as university libraries and saying, "Hey, don't you want a subscription to the Appalachian Journal of Holistic Particle Physics and Macrame?" If I was a library, I'd probably come back with, "No, I don't think we're likely to have a lot of researchers who need articles from the AJHPPM. If we ever did have someone who needed an article from that journal, they could just buy the individual article." The publishers don't like that answer, so they would like buying individual articles to be prohibitively expensive.
Of course Nature isn't in the same league as AJHPPM, but that just raises the question of why the per-article prices seem to be so uniform across publishers and journals. For example, nobody seems to be experimenting with discount pricing. I suspect that this is an example of price fixing.