Adding more worker agents makes the system slower

• A supervisor delegates to specialist worker agents. Going from 3 to 8 workers increased end-to-end latency.
• The workers spend most of their time idle, waiting.
• The supervisor is at near-100% utilization while workers sit at ~20%.

topology: 1 supervisor -> {8 workers}
supervisor: serializes every task, calls workers one at a time, re-plans after each result
worker utilization: ~18% avg   supervisor utilization: ~96%
tasks are independent (no data dependency between them)