Beyond 60% utilization, the average response time doubles. By 80%, it has tripled. As disk I/O latency is often the bounding resource for an application, increasing the average latency by double or higher can have a significant negative effect on application performance. This is why disk utilization can become a problem well before it reaches 100%, as it is a queueing system where requests (typically) cannot be interrupted and must wait their turn.

— Brendan Gregg, Systems Performance