For cutting edge theoretical research, there are only a handful of people in the world with the knowledge to review the research. And having millions of people that really don't have a clue about it doesn't produce anything worthwhile.
Yes, but those who carry out the peer-review, are the researchers themselves: at times, at little to no cost to the publisher, simply because the reviewing scientist (and the postdoc she/he usually assigns to do the grunt work) get the "prestige" of being a reviewer/editor for a given publication.
There seem to be two issues here that are being conflated.
The first is the value of peer review. I don't agree with the arguments that having a select group of people, who are experts in their fields, reviewing papers is a bad thing. Nor do I think that opening it up will result in anything other than a terrible amount of noise.
The second argument is over the necessity of for-pay journals. Here, I think there could be a lot of work... if the new journal still provides the same amount of review and scrutiny as the current ones.
So it is a valuable service to provide.