Server
performance.

This paper evaluates Gray as an integrated language and server runtime under corrected-latency HTTP load. A digest-pinned, one-CPU comparison exercised equivalent plaintext and native-JSON handlers in Gray Static, Go, Node.js, and PHP across seven requested rates and retained all 56 runtime-route-rate points. Under a declared controlled-delivery criterion of at least 95 percent rate delivery with corrected p99 below 100 ms, Gray's highest point was 60,000 requested requests per second for both routes; Go's was 30,000 for both, Node.js reached 12,000 for plaintext and 20,000 for JSON, and the nginx plus PHP-FPM fixture reached 8,000 for both. Gray delivered 59,259.63 plaintext and 59,257.18 JSON requests per second at the 60,000 points, then visibly crossed its own queueing knee at 75,000. A separate sequential qualification sustained plaintext at 49,997.81/s and JSON at 39,993.57/s for one hour each with corrected p99 of 7.00 ms and 4.09 ms and no sampled steady-state RSS or descriptor growth. The database phase sustained 3,999.85/s but recorded 802.82 ms corrected p99, exposing a blocking wait architecture subsequently replaced by compiler-hidden continuations. The combined record demonstrates that Gray's integrated compiler-runtime path achieved a materially later controlled throughput knee than the compared fixtures on this host while preserving exact-response, resource, and failure gates.