docs: add benchmark page and performance references

Author: cshum

docs/docs/benchmarks.mdx

@@ -6,7 +6,11 @@ description: Benchmark overview for imagor, including methodology, measured resu
 
 imagor is one of the fastest image processing servers available today.
 
-That result comes from the combination of [libvips](https://github.com/libvips/libvips), Go binding [vipsgen](https://github.com/cshum/vipsgen), and imagor's streamed processing path. imagor keeps the request path close to libvips, avoids unnecessary buffering and copies for common streamed inputs, and exposes a libvips-friendly source path that works well for resize and thumbnail workloads.
+That comes from the combination of [libvips](https://github.com/libvips/libvips), the Go binding [vipsgen](https://github.com/cshum/vipsgen), and imagor's streamed processing path. imagor keeps the request path close to libvips, avoids unnecessary buffering and copies for common streamed inputs, and exposes a libvips-friendly source path that works well for resize and thumbnail workloads.
+
+Last updated: May 2026.
+
+This page is based on recent reruns from the public [image-servers-benchmark](https://github.com/cshum/image-servers-benchmark) project, using current released versions and the benchmark summaries linked below.
 
 ## Why imagor is fast
 
@@ -95,6 +99,7 @@ Benchmark shape:
 - AWS EC2 `c7i.large`
 - `2 VUs`, `5m`, `10s` warmup
 - resize to `512x512`
+- public `image-servers-benchmark` project with the rerun configuration documented in the linked benchmark summaries
 - compared runtimes:
   - imagor: `ghcr.io/cshum/imagor:1.9.1`
   - imgproxy: benchmark harness used `darthsim/imgproxy:latest`, corresponding to imgproxy `v4.0.3`