Why would a browser use HEAD instead of conditional GET to re-request cached content? It would require one more round-trip if a refresh is actually required.
This acts like HEAD if the resource has not been modified and like GET if it has.
In practice this seems to be the go-to technique for browsers as well.
I looked it up, though. It looks like you have to have an ETag to do a conditional GET, so browsers use HEAD if the original response (the cached one) didn't include one.
Nope, not the case. May I ask where you read that? It differs from my reading of RFC2616 and my experience.
If the server gave you an ETag though you're supposed to include it in future cache-conditional requests. Section 13.3.4 (http://www.w3.org/Protocols/rfc2616/rfc2616-sec13.html#sec13...) describes the interaction between ETags and modification dates.
Am I correctly inferring that you need at least a Last-Modified in the response to do a cache-conditional request, though?
Edit: The difference is that Last-Modified allows clients to use a heuristic to determine if the response should be cached for a certain duration (unless explicit Cache-Control or Expires headers are used). The heuristic isn't specified, but a 10% fraction of Date - Last-Modified is suggested, and in fact, that is what e.g. Internet Explorer does.