> They also manage the review process (screen papers, find reviewers, communicate with authors and reviewers),
They do not. That's the editor's job, and Elsevier journal editors are unpaid volunteers -- as the Guardian article correctly pointed out.
> A half-decent review can easily take 3-4 hours, and a typical paper gets 2-3 reviews, so something like 8-12 hours of PhD-level labor which are not currently compensated.
It generally takes much much longer than that.
The average article in Neuroimage is 14 pages long (~ 10000 words), not counting references and declarations. Pure reading, single-domain, technical: the chart says 10k words will take at least 60 minutes. And you'll definitely need to read the article more than once while doing peer review.
Then you'll need to look up references, do some sanity check calculations, evaluate artifacts if available, scrutinize figures, and unless you're up-to-date on all recent developments of the field, probably read at least one more article on prior work just to understand what's going on. If unusual methods were used in the evaluations, you have to understand these too, so better add a few more days.
And you still have to actually write the review: at that point, you only have some notes and scribbles! In neuroscience, we're talking about at least 2-3 days of FTE work. And that's only the first round of reviews. About 80% of articles go through multiple rounds of review-response, according to Neuroimage's own statistics.
In other fields, review times might be significantly longer (weeks, or in mathematics even months, instead of days).
Elsevier profit margins are 40%, according to their own admission. They could afford the costs of compensating this labor. But why would they?