Increasing concurrency in HTTP clients doesn't just raise the number of in-flight requests — it also creates a burst of simultaneous request starts at t=0 and unpredictable per-host concurrency in multi-host workloads. Both hidden factors can trigger HTTP 429 rate-limit errors even when your global pool size seems reasonable. Benchmarks show that adding a start-rate cap (independent of pool size) can recover useful throughput from 16/100 to 100/100 successful requests at the same concurrency level. For multi-host workloads, a per-host limiter nested inside the global pool prevents any single strict host from absorbing a disproportionate burst. Residential proxies are also discussed as a workaround that raises the effective burst tolerance without fixing the root cause. A practical checklist covers measuring observed start RPS, per-host peak in-flight counts, and optimizing for successful requests rather than raw completed request rate.
