when does a message queue become a bottleneck?

We've been using a message queue to decouple services, and it's worked wonders for throughput and resilience. But, I'm starting to wonder about the tipping point. When does the queue itself, or the producers and consumers around it, become the limiting factor? Are there specific patterns or metrics you look for to identify when your message queue is struggling or causing downstream issues? I'm really curious to hear your experiences and strategies for optimizing or scaling them before they become a major pain point.