In a recent meta.physics.se discussion on the use of permanent links form physics.stackexchange, I proposed as policy that
when linking to the arXiv, link to the abstract page (arxiv.org/abs/...) and not directly to the pdf, as a courtesy to users on lower bandwidths. If the paper has been published (which will be noted on the arXiv abstract), include a link to the journal version, even if it's paywalled, if only as a courtesy to the referees' hard work.
One comment on the italicized part was
Doesn't pretty much everybody post the final version on arxiv?
and I would like to know (if that is even possible) to what extent this is true. My gut feeling is that this is field-dependent, but I do not know what evidence may be available to explore this, other than anecdotal evidence.
Thus: Is there a clear trend for authors to upload post-referee-process versions after publication? Does this depend on the specific field? What evidence is available for this?