The admission that all 5,000 sites are identical clones matters more than it looks. Identical content means one shared working set in the OS page cache, so this

The admission that all 5,000 sites are identical clones matters more than it looks. Identical content means one shared working set in the OS page cache, so this mostly measures template rendering, not the memory behaviour of 5,000 different datasets — which is the thing that actually decides whether a small box survives. The number I'd stare at is p95 = 1.43s at 146 rps with CPU peaking at 81%. That's the point where a small VPS stops being comfortably idle and becomes one background job away from trouble. We run a much smaller site on a similar box, and what breaks latency there has never been traffic — it's the crawler and image generation competing for the same two cores. Is that server serving only, or does it also run the background work for those sites?