You may want to check out this thread: API rate limiting and Streaming API endpoints - clarification
In general if you’re using multiple keys on a single server and their cumulative requests exceed the rate limit you’re at risk of being banned at the IP level. As such you may need to build some form of internal rate limiting (or queueing). That said, we have a system like this in place, and despite having a large and active user base it’s never actually been triggered. I think most use cases where this was going to be an issue you’d probably be better off using the bulk data-sets.