I suspect from conversations with journal editors that it's not a whole lot cheaper to produce CS/math journals than other types, but that's just a general impression, not a certainty.
There seem to at least be some that are much cheaper, in the sense of having budgets literally approaching $0. From some brief chats with JMLR editors, they're in that camp: they run on donated server space from MIT and volunteer editors. No staff, no office, no recurring expenses, yet they're still one of the top journals in CS. That seems like a significantly different cost structure from something like PLoS ONE, perhaps closer to the arXiv model turned into a journal.
There seem to at least be some that are much cheaper, in the sense of having budgets literally approaching $0. From some brief chats with JMLR editors, they're in that camp: they run on donated server space from MIT and volunteer editors. No staff, no office, no recurring expenses, yet they're still one of the top journals in CS. That seems like a significantly different cost structure from something like PLoS ONE, perhaps closer to the arXiv model turned into a journal.