Read the measured comparison
At a target of 50 requests per second, three rotated runs on one disposable host produced these medians. All three actually issued approximately 50 requests/s; all recorded requests returned HTTP 200. WordPress memory includes MariaDB.
| Deployment | Median run p95, ms | Median sampled peak working set, MiB |
|---|---|---|
| GoRun | 2.66 | 114.54 |
| WordPress + MariaDB | 16.85 | 203.78 |
| Astro static + Nginx | 0.66 | 4.09 |
What this measured
The fixture served 100 identical synthetic articles with minimal templates and equal 4,513-byte responses, using five seconds of warmup and 30 seconds of measurement per run. Each deployment had two logical server CPUs (SMT siblings of one physical core on this runner) and a 2 GiB total budget. There was no TLS, CDN, full-page cache or plugin workload. Peak working set is sampled aggregate container memory minus inactive file cache, not process RSS or an exact continuous peak.
GoRun used 0.4.0-era application source, not the 0.5 UI. WordPress was 7.1.3 with PHP 8.3 and MariaDB 11.4; Astro was 7.4.0 served by Nginx. This WordPress configuration was not presented as tuned. Exact source/image identities and limits are in the evidence.
Read the higher target carefully
At the nominal 200 requests/s target, the generator’s 100-worker cap delivered only a median 91.74 issued requests/s to WordPress, with 90.24 successful requests/s after final-response waiting. That condition is not an equal-200-load comparison or a maximum-capacity result. Issued rate and successful throughput use different time intervals; configured target is not delivered traffic.
What you can conclude
GoRun had lower latency and sampled working set than this uncached WordPress setup in this workload. Astro’s prebuilt static pages were lowest on both measures; Astro is a static-delivery reference, not an equivalent full CMS. These measurements do not compare editing, installation, model quality or complete product capability. Edge has no runnable CMS here and no measured result. Use the linked raw archive and executed notebook to check or reproduce the figures.