Just to clarify: we count both internal (from workflows, among others) and workspace repos to the 5K count. For workflows where the repos count are exceeded, execution is blocked initially for 10 minutes until the count is reduced. There is a cleanup process for finished tasks as well.

Does the job eventually get executed, or did it completely fail?

This is why it's important to get a complete repos count so we can check if this is the behaviour that you are seeing.