20 years ago, the process involved the creation of a physical artifact -- bound paper, handed out to participants, mailed to subscribers, etc. This is a fairly non-trivial undertaking, certainly costing quite a few thousand dollars.
As of a few years ago, the last of the major conferences in my area stopped creating printed proceedings. Along with that goes away the need (desire?) to ensure precise formatting rules, etc. -- the job of the typesetter is gone. We just submit PDFs, and everyone does their own.
The other "costs" of a journal or conference were already borne by the research community -- most of our conferences and journals are all-volunteer. It's a decentralized cost spread across the salaries of everyone in the community.
In our model, at least, the role of the publisher is greatly diminished. They provide some hosting, some indexing... and not much else.
Not all fields are like that. Science and Nature, for example, have paid editors. Fine. Happy to pay for that.
The shift in costs happened a while ago - the costs are probably in the range of $10-$100 per paper now, compared to quite a bit more before.
Not necessarily. In my own field (a branch of linguistics), many of our journals are open access, but they are still free to publish in. Some of the publishing costs are covered by the fact that these journals are put out by learned societies that are sitting on large endowments or receive state funding, and even for journals that are open access online, libraries often still want to pay for hard copies.
Hindawi is half that, but they will publish anything.
You've misunderstood something there. With the "open access" model the costs are merely shifted around. In the traditional model the library would pay for subscription, now it's the researcher paying page charges for publication. In the past of society publishers the society had a reputation to lose, but Hindawi doesn't have such restraints on it.