Our AEM dispatcher cache becomes invalid as soon as any page is (re)published by an author. As we publish pages many times each hour, there is effectively little to no dispatcher page caching, and pages are rebuilt for each request. I can see the effect of this with the marked difference in response time during a period of no author publishing and just after.
Is this expected? Can't the cache be configured so that only the re-published page and those dependent on it are invalidated?
The dispatcher allow you to serve a very large number of request if the page is cached.
If you need to remove all cache for each author page publish, probably you are working on a full loggedin site which need to have really dynamic information. If you are working on a full loggedin site, probably you need to remove the caching directly from the dispatcher configuration; even if I can confirm you that also into some loggedin site (e.g. intranet) something could be cached in order to improve performance.
You might want to revisit the dispatcher cache configuration and check statfileslevel. Having said so , you can also Cache content for as long as possible and reduce requests that go back to AEM - Optimize flush requests by enabling refetching flush on all flush agents. Use /enableTTL and set Cache-Control: max-age=... header to cache files as long as possible.