[Bromskloss]

> even the researcher who wrote it will have to pay.

Eh, pay for what? The author already has the article. Often-times, the agreement furthermore permits limited dissemination of the article by the author, possibly even on the author's web page.

[diffeomorphism]

Pay for color figures in print, for example (a few hundred per figure if I recall correctly).

Furthermore, you get an author's copy with watermarks with very limited dissemination rights. You don't get a copy of the journal issue unless you pay.

Also, this varies very much by discipline. Math research is pretty universally available on the arXiv, but chemistry or biology is not nearly as pervasive.