I can't repro.  When you say "visit multiple times", you mean navigate to that URL multiple times, not reload it, right?  Reloading will force cache-revalidation.  Just selecting the omnibox and pressing enter without modifying the URL will also be treated as a reload instead of as a navigation.

No, I meant either pressing enter in the omnibox or F5. So, does that mean it is by design to ignore the cache in this situation, even if the cache headers previously said immutable and cache for a year?

The original use case demonstrating it is inside an encrypted file storage app (Peergos) which makes ajax requests to similar urls with the same etag and cache-control header behaviour. I'll see if I can put together a public repro. 

Yes, that's by design.  End users generally press F5 or press enter in the omnibox when there's a problem, so we re-validate the main resource in that case (But not subresources, though we used to revalidate them, too).

Most other cases will use the cache (Fresh navigation, bookmark, link, etc).  So it sounds like this is working as intended.

Thank you for clarifying the design. Feel free to close this issue.

Done, thanks for the followup!