How does this differ from using a templating engine where the server gets a template, and instantiates it to "the final thing it should be" on a per-request bas

How does this differ from using a templating engine where the server gets a template, and instantiates it to "the final thing it should be" on a per-request basis? (also, if you need to transform 200 requests per second that's slowing things down, there's almost certainly a cache-layer missing?)