A company deploys a text summarization service using a large language model on a cluster of four identical devices. They use a simple 'round-robin' scheduler, which sends each new request to the next device in a fixed cycle. The service accepts documents of widely varying lengths for summarization. System monitoring reveals that during peak usage, average response times are high, and device utilization is highly imbalanced—some devices are constantly busy while others are often idle. Based on this scenario, analyze the root cause of the performance issues.