Also, all of my colleagues would laugh at me if I published in a junk venue. There's no win, at least in the top tier.
That's the heart of the problem right there. Would you laugh your colleagues if they published in a junk venue?
Or would you look at the work on its merits and ignore the medium?
I wouldn't do anything of the sort if they published on arxiv. We have a place to publish respectable work without fees and without peer-review. In some subfields of CS, we're also lucky enough to have high quality, open access venues without exorbitant fees (e.g., USENIX conferences). So pay-for-crap is not a necessary choice.
That's the whole point of this debate. All the paid access journals are leeches in one form or another.
Additionally, to get at your earlier question of evaluating a paper by its host journal: Because academics are strongly incentivized to publish in high-impact journals, if someone publishes in a crap journal it raises eyebrows, because the obvious question is 'why' and the obvious answer is 'because the authors didn't want rigorous review' or 'the work is of supreme unimportance'.
Furthermore, because of curation aside from peer review (editors weeding out solid but unimportant papers) the journal's brand makes a difference in how many people will actually see the paper. While many readers will come across papers either by specific searches in Google Scholar, or by going through the citations in a different paper, many of us still read the tables of contents of a few key journals every week or month to see what is being produced.
I'm a non-academic scientist (in the non-profit world currently) and though I will never face a promotion and tenure committee who really gets into impact factors etc., I still want my work to be as widely read as possible, so I push for the best journals that I can get in that still meet my criteria (non-profit societies that have open-access at least after a year embargo). Google Scholar also ranks papers by how highly cited they are, so there are still viral-type effects and getting eyeballs on a paper early greatly affects later discovery and engagement.
edit: I took 'paid-access' to mean 'pay to publish' which was the original thread's wording, and wrote based on that. Some of this reply still holds for 'pay to read' but not all.