There are a couple of background issues worth addressing before actually answering your main question. Also what I say is I think fairly typical for math and theoretical CS; I guess some but not all is similar for other fields.
You’ll get faster at reading papers as you gain experience. Much, much faster. I remember in my first couple of years of grad school feeling similarly to you: reading a paper was a big undertaking — it could take weeks of work to get through a particular paper. Now that’s very rare — even a pretty thorough read of a paper (e.g. for a review) is typically under a day’s work. Longer than that comes only occasionally, if I need to really thoroughly understand an unusually substantial and novel paper. Mostly, it’s just being more familiar with the landscape the papers are working in, and the techniques used. Also, partly, reading papers is itself a learned skill.
You’ll learn how to read a paper in less than full detail (and when that’s appropriate). Again, I remember first reading papers, I would go through them line by line and make sure I absolutely understood every step. Now, I usually don’t need to, for multiple reasons. One is that I no longer need all the details in order to follow the general outline, because I understand the big picture better. Another is that I’m better at judging in advance which details I will or won’t need to know: e.g. if it looks like some proof is a routine calculation, then I’ll skip it, knowing that I can come back to it later if I ever need a similar calculation and have trouble doing it myself. Another is that I may be able to see in advance (e.g. from the introduction) that some sections will be more relevant to my work than others. And, of course, I can skip or skim parts that recall background material that I already know.
So, coming back now to answer your main question: The actual proportions will vary greatly, but typically, authors will have read all the cited papers at least in part, but will not have read most of them in full detail. That said, reading 12–15 papers thoroughly isn’t infeasible by any means.
Looking at the bibliography a recent 2-author paper of mine with about 30 references in the bibliography, they break down very roughly as:
30%: I’ve read these papers/books carefully in their entirety at some point in the past.
40%: I’ve read parts of these papers/books thoroughly (including the parts relevant to what we’re citing them for), and at least skimmed all or most of the rest.
20%: I’ve just dipped into these papers/books for specific results we needed to cite (plus relevant background); I’ve not read most of the rest.
10%: I’ve not looked seriously at these ones at all; my co-author is responsible for the citations to them.
This is I think fairly typical for me; but I also know some authors for whom I suspect those numbers would typically look more like 10-30-40-20, and others for whom they’d be more like 80–20–0–0.