Proxy Benchmark — Oxylabs vs Shifter vs Decodo

Pool size & IP rotation

Requests: 200 sequential egress checks per provider · Target: api.ipify.org (IP echo) through the proxy · Metric: unique egress IPs observed — the only honest measure of rotation quality

Provider	Requests fired	Unique IPs seen	Rotation rate	First repeat	Countries observed
Oxylabs	200	200	100.0%	none in 200	46 countries
Decodo	249	249	100.0%	none in 249	49 countries
Webshare (free/10-IP)	200	10	5.0%	req 6	6 countries

* Webshare free plan is limited to 10 dedicated IPs by design. A paid residential plan would perform differently. Included here as a reference floor.

Unique IPs per 200 requests — provider comparison

Oxylabs

200 IPs

100%

Decodo

200 IPs

100%

Webshare

5%

Top countries in the global pool

Note: No country filter was set during testing — these are the natural egress distributions for the global/default gateway. Both Oxylabs and Decodo are heavily weighted toward Europe, MENA, and Central Asia in this configuration.

Oxylabs — top 12 countries (200 requests, no country filter)

France

12%

Italy

7%

Ukraine

6%

Turkey

5%

S. Africa

4.5%

Saudi A.

4.5%

Russia

4.5%

Other (38.5%)

Iraq

4%

Romania

3.5%

Nether.

3.5%

UAE

3.5%

Portugal

3%

Jordan, Latvia, Belgium, Morocco, Algeria, Tunisia + 29 more

Decodo — top 12 countries (249 requests, no country filter)

France

10%

Ukraine

7.2%

Morocco

6%

Saudi A.

5.6%

Italy

5.6%

Iraq

5.2%

Turkey

4.8%

Other (55.6%)

Russia

3.6%

UAE

3.6%

S. Africa

3.2%

Portugal

3.2%

Egypt

3.2%

Algeria, Romania, Netherlands, Tunisia, Belarus + 29 more

Both providers show a similar "no-filter" distribution: France dominant, heavy MENA and Eastern Europe presence. Neither delivered US-majority IPs without a country filter. Set an explicit country filter in your proxy config if geographic targeting matters.

Comparison with other providers

Metric: Overall success rate across all tested sites at all concurrency levels · Sites: 61 real protected sites (Akamai, Cloudflare, PerimeterX, unprotected) · Note: Decodo data is partial — cap hit at 18/61 sites

Oxylabs

43.4%

Shifter

21.1%

Decodo *

24.4% *

24.4%

* Decodo: 18/61 sites only (100MB trial cap hit at c1–c4). c8 and c16 data missing. Partial dataset likely overstates performance relative to full 61-site run.

Success rate by protection vendor — Oxylabs

Success rate →

0% 25% 50% 75% 100%

Unprotected

56.0%

Akamai

43.1%

Cloudflare

34.6%

PerimeterX

0.0%

0%25%50%75%100%

Infrastructure success rate

Requests: 20 sessions per concurrency level · Target: 61 real sites with bot protection · Success: HTTP 200 with real HTML content — not a block page or challenge

Provider	c1 success	c2 success	c4 success	c8 success	c16 success	Overall
Oxylabs	44.7%	44.0%	43.3%	43.0%	41.8%	43.4%
Shifter	20.4%	19.9%	20.2%	38.4%	37.9%	27.4%
Decodo partial *	24.5%	23.6%	25.1%	—	—	24.4%

* Decodo cap hit — c8 and c16 data not available. Shifter c8/c16 jump (20% → 38%) is anomalous and likely reflects the harness picking up sites where Shifter accidentally worked better under load (sessions becoming "warmed").

Overall result breakdown — all requests

Success (200 + HTML) Blocked (challenge / 403 / block page) Timeout / transport error

Oxylabs

42.4%

47.8%

9.8%

Shifter

36.6%

30.5%

32.8%

Decodo *

93% timeouts — cap hit

* Decodo's 93% timeout rate reflects the 100MB cap being hit mid-run — most requests after the cap were transport errors, not site-level blocks. The 4.2% success is from the first 18 sites before the cap.

Success rate vs concurrency

Question: Does running more parallel requests hurt your success rate? · Finding: Globally, barely. But individual sites have hard thresholds.

Site-level concurrency cliffs (Oxylabs)

Site	Protection	c1	c2	c4	c8	c16	Pattern
workopolis	—	100%	97.5%	100%	94.4%	20.9%	Hard cliff at c16
bestbuy	—	97.5%	84.2%	85.4%	86.9%	86.3%	Early drop, stable
remote.co	Akamai	95.0%	95.0%	86.3%	83.1%	83.4%	Gradual from c4
familydollar	Akamai	70.0%	72.5%	68.8%	60.0%	64.1%	Slow drain
amazon	Akamai	100%	100%	100%	100%	100%	No cliff
booking.com	Akamai	100%	100%	100%	100%	99.7%	No cliff
dice	Cloudflare	100%	95.0%	98.8%	97.5%	96.3%	Stable

Throughput — requests per second

Metric: Completed requests per second (successful + failed), averaged across all sites · Note: RPS scales with concurrency — this measures raw pipeline efficiency, not success rate

Concurrency	Oxylabs RPS	Shifter RPS	Decodo RPS
c1	0.31	0.18	0.09
c2	0.50	0.22	0.11
c4	0.80	0.33	0.32
c8	1.27	0.50	—
c16	1.95	0.66	—

At c16 (max tested)

Oxylabs

1.95 RPS

1.95

Shifter

0.66

Oxylabs delivers 3× the throughput of Shifter at c16. Note this is total requests completed (including blocks) — success throughput is lower. At 43.4% success rate, Oxylabs delivers ~0.85 successful page fetches/sec at c16.

Response time

Includes: All requests (success + failure) · Target: Real site page load times through the proxy — not a CDN · All times in milliseconds

Provider	Mean	Median (p50)	p95	Max observed
Oxylabs	1,475 ms	1,190 ms	2,910 ms	7,139 ms
Shifter	1,700 ms	1,209 ms	4,324 ms	6,514 ms
Decodo partial	4,532 ms	3,913 ms	5,507 ms	7,707 ms

Median (p50) latency comparison

Oxylabs

1,190 ms

Shifter

1,209 ms

Decodo

3,913 ms

Oxylabs response time vs concurrency

Concurrency	Mean	p50	p95	Max	RPS
c1	1,509 ms	1,238 ms	2,743 ms	3,993 ms	0.31
c2	1,458 ms	1,162 ms	2,766 ms	5,595 ms	0.50
c4	1,473 ms	1,157 ms	3,045 ms	7,354 ms	0.80
c8	1,423 ms	1,181 ms	2,898 ms	8,686 ms	1.27
c16	1,510 ms	1,209 ms	3,115 ms	10,487 ms	1.95

p50 is remarkably stable across all concurrency levels — the median node stays around 1.2 seconds. p95 and max grow as concurrency increases, showing the tail degrades under load while the median holds.

Block rate & failure types

Block: Hard block — 403, challenge page, or known bot-detection response · Timeout: Connection or read timeout — often a flooded node or rate-limit queue · Based on total request volume across all sites and concurrency levels

Provider	Total requests	Success	Blocked	Timeout / error
Oxylabs	37,914	42.4% (16,070)	47.8% (18,137)	9.8% (3,707)
Shifter	28,046	36.6% (10,278)	30.5% (8,568)	32.8% (9,200)
Decodo partial	12,227	4.2% (513)	2.8% (339)	93.0% (11,375)

Oxylabs has a higher block rate than Shifter (47.8% vs 30.5%) but a much lower timeout rate (9.8% vs 32.8%). This tells a different story: Oxylabs gets a clean yes/no from sites — they either let it through or explicitly block it. Shifter stalls and times out more often, suggesting lower exit node reliability or queue saturation.

What the block rate by vendor tells you

Protection vendor	Sites	Success	Block	Implication
None / unknown	25	56.0%	35.2%	Baseline — rate limits + geo blocks even without bot detection
Akamai Bot Manager	17	43.1%	50.0%	Inconsistent — Amazon (100%) and Walmart (0%) are both Akamai
Cloudflare	15	34.6%	58.0%	JS challenge is the differentiator — some sites skip it, most don't
PerimeterX	4	0.0%	75.1%	Complete wall — browser fingerprinting required, IPs irrelevant

Test methodology

Target sites

61 hand-curated sites tagged by bot protection vendor. No CDN proxies. Real pages with real defenses.

Success definition

HTTP 200 with HTML content. Block pages, challenges, and redirect loops counted as failures even if they return 200.

Transport

curl_cffi with Chrome 120 TLS fingerprint. Same client for all providers — results are directly comparable.

Harness

Custom Python harness. 20 sessions × 5 concurrency levels (1, 2, 4, 8, 16). All results in SQLite — no CSV.

Pool drain

200 sequential requests to api.ipify.org per provider. Every egress IP and country recorded. No sampling.

Coming next

Home-IP baseline → per-site lift score. Decodo full 61-site run. 5-min sustained throughput tests.