At least in my field (applied machine learning/optics) and my partner's field (cognitive neuroscience) pre-prints are a massive deal. Not only do they allow you to get your work out earlier, google scholar will typically link your pre-print to the published article eventually so anyone who doesn't have a subscription can get a PDF no problem. This probably nullifies the negative effects of not having access to the publisher's library.
It's to the point now I'd say that all of the grad students and younger PIs insist on posting a pre-print. Pretty great trend especially considering with LaTeX the preprint ends up looking just as good as the published version anyways....