This has actually succeeded before though. Journal of Machine Learning Research (JMLR) is one of the premier journals in its field, is open access, and was formed by the mass resignation of editorial board of 'Machine Learning'.
Yes, and ‘Machine Learning’ is still being published, as is virtually every other journal of note that’s seen a mass editorial resignation. Maybe this strategy will work in the long run—if all the high-prestige journals are open access the publishers may wither and die—but so far Springer and Elsevier and the others make their money either way.
JMLR has a far higher impact factor than "Machine Learning".
The for-profit publishers are welcome to print whatever they want, forever. They even publish "hollow" journals that exist purely as spoilers. You're mistaken if you believe they "make their money either way" they need subscribers, and it's increasingly unattractive to subscribe to second rate journals like "Machine Learning".
IIRC, the way they defeat this is by offering discounts on subscriptions to large groups of journals, rather than to individual titles.
If you're going to have to buy the package anyway, any money that is spent on journals not in the package is an extra expense that will have to be individually justified.
There’s a number of these stories, such as the former board of (now-dead) Journal of Algorithms founding the ACM Transactions of Algorithms and the former board of (now-dead) Topology founding the Journal of Topology.
https://en.wikipedia.org/wiki/Journal_of_Machine_Learning_Re...