But if the queue would only be per node the whole point of throttling would be gone or not? At least I would have expected that you want to avoid something like hitting some kind of API to fast. If it would not work on a global level for all workflows that run at the same time it would be useless.
It would only work for if there is a loop in a workflow and if it has a lot of items and the nodes afterward has to make a separate API request for each.