What is a good success rate for residential proxies?

Industry-published success rates for reputable residential proxy networks typically range from 95% to 99.9%. In this 10,000-request test, Roam's rotating residential network delivered 99.67% overall, with the lowest single-country figure at 99.30% (Brazil).

How fast are residential proxies compared to datacenter proxies?

Residential proxies are slower because traffic exits through real consumer connections. In this test the median full-request time (DNS, TLS, response over HTTPS) was 2.0–2.7 seconds depending on country, versus roughly 0.2–0.5 seconds typical for datacenter proxies.

How was geo-targeting accuracy measured?

Each request pinned one target country, and the response reported the country the request actually exited from. Accuracy was 98.9%–100% per country: every request to Japan and Brazil exited in the right country; the worst case was Germany at 98.9%.

Residential Proxy Benchmark: 10,000 Requests Tested (July 2026)

Test run: 10 July 2026 · Published: 10 July 2026 · Raw data: CSV, 10,000 rows (CC BY 4.0) · We update this page whenever we re-run the test.

TL;DR: We sent 10,000 real HTTPS requests through Roam's rotating residential proxy network — 2,000 each targeting the US, UK, Germany, Japan and Brazil, forcing a fresh session on every request. The network answered 99.67% of them successfully, used 7,436 unique exit IPs, hit the requested country 98.9–100% of the time, and delivered a median full-request time of 2.0–2.7 seconds per country.

What exactly did we measure?

Every data point below is one complete HTTPS request — DNS resolution, proxy CONNECT, TLS handshake and response body — issued with curl through our production gateway (gw.roamproxy.com:41080, measured from the same host to exclude client-side network noise). A request counts as successful only if it returned HTTP 200 within a 30-second timeout. Nothing was retried; every failure stays in the data.

Results by country

Target country	Requests	Success	Median	p90	p95	Unique exit IPs	Geo accuracy
United States	2,000	99.80%	2.00 s	5.70 s	7.27 s	1,827	99.5%
United Kingdom	2,000	99.80%	2.37 s	7.31 s	8.23 s	1,533	99.4%
Germany	2,000	99.75%	2.39 s	7.56 s	12.22 s	1,651	98.9%
Japan	2,000	99.70%	2.03 s	11.09 s	14.01 s	1,026	100%
Brazil	2,000	99.30%	2.73 s	12.46 s	17.18 s	1,399	100%
All	10,000	99.67%	—	—	—	7,436	—

Reading the numbers: median times cluster tightly at 2.0–2.7 seconds everywhere, while tail latency (p95) stretches with distance and market — Brazil's consumer connections are simply more variable than US ones. That long tail, not the median, is what you should budget for when setting client timeouts; we suggest 20–30 seconds for residential traffic.

How diverse is the IP pool really?

Because each of the 10,000 requests forced a new session, the test samples how often the network hands you a fresh address. It returned 7,436 distinct exit IPs — 74% of requests got an IP no other request in the test saw. The US sample alone touched 1,827 unique IPs in 2,000 requests (91%). Between 42% and 76% of exits were IPv6, which is typical of real consumer last-mile connections.

Methodology (reproduce it yourself)

Date: 10 July 2026, single continuous run.
Volume: 10,000 requests = 5 countries × 2,000; concurrency 20; no retries.
Rotation: username-pinned country targeting with a unique sticky-session tag per request (-country-us-session-N), so every request asks the pool for a session rather than reusing one.
Target: https://www.cloudflare.com/cdn-cgi/trace — small, globally anycast, and it echoes the exit IP and country, which is how geo accuracy was scored.
Vantage point: a server in Tokyo, Japan (AS3258). Latency from your location will differ; success rate and pool diversity should not.
Tooling: plain curl with time_total / time_starttransfer timers; per-request rows in the published CSV (exit IPs anonymized to /24 and /48 to protect residential users).

Honest limitations

One vantage point (Tokyo) and one small target endpoint. A heavy page on a slow origin will take longer than these figures; a benchmark measures the network, not your target site.
Success against Cloudflare's trace endpoint does not guarantee success against sites with aggressive bot defenses — that depends on your client fingerprint as much as the IP.
This is our own network measured by us. We publish the raw data and the exact method so you can verify rather than trust — running the same loop against any competitor takes one afternoon.

Context: how does this compare?

Independent testers such as Proxyway typically report 95–99.9% success rates for the major residential networks, with average response times between 1 and 3 seconds measured from US/EU vantage points. Roam's 99.67% sits at the upper end of that success band; our medians (2.0–2.7 s from Tokyo) are in the same range once vantage distance is accounted for. Pricing context is on our comparison page, verified against official pricing pages in July 2026.

Try the same network — $2/GB metered, no subscription. Or read who runs Roam.