It’s usually not images but rather the CMS making 100 read queries and 20 write queries per page load without any object caching that brings sites down. Even a slow uplink serving big image won’t bring an nginx server down, it’s all async.

Serving a lot of images at the same time will work the disk I/O up in many cases. Probably gotten a lot better with NVMe though.

It’s typically one page that hits Reddit/HN/whatever with its fixed set of static images. It’s pretty much the ideal scenario for the kernel’s in-memory cache.