Hi Andre, I'm not sure what might be causing this. Are your deployments using ECS infra blocks, or are flows running directly on the agent?
First, I'd check to see how the resource metrics are looking on your agent, it may be under-resourced depending on the volume of flow runs it's handling. I'd also take a look at the concurrency limit settings everywhere they exist to ensure nothing is misconfigured. You may also want to investigate upgrading your agent to a worker.