In practice, for maths papers I look on arXiv, for crypto papers on iacr and so on - academics generally want their work to be read (and cited) so they're usually happy to make it available for free. There's even tricks you can play like uploading an "author version" or "preprint" if you're forced to use a commercial publisher for a conference.
That also means that if the whole racket is revisited at some point, then Elsevier will get to pick on someone their own size if not bigger - and will hopefully come off worse in that fight.
What? Elsevier doesn't pay reviewers either.
And it's not even the authors fault the system is like that. Research isn't just about saying "hey we found this works", but also about "we wasted 3 years, it doesn't work sadly". Yet, the second option does not lead to the same impact, because if something doesn't work it's not going to be reproduced and thus quoted.
You're not gonna be published on high impact journals with data that doesn't move the field, even though as you point out the information is as valuable.
a less extreme widening of scope would be just the research funds that apply or deny grants for research: suppose a flurry of papers investigates the superconductive behavior of a piece of meteorite, and you're the one to kill the buzz with your negative result, then future grant denials could cite your boring negative result.
So the publisher still gets a significant amount of money, albeit often paid by the institution of the author.