The only guesses I have are:
1 - some form of server DOS protection that is holding down the connections from the SF agent
2 - some really screwy routing between the SF agent and your server (though I’d expect that to show up in the socket connect time)
3 - a problem with the test agent that only manifests itself on your site
It looks like the server ACK’d the actual HTTP request and then 20 seconds later the response started coming so from a network perspective it looks fine (but slow).
If you don’t already have it turned on, you can add the server response time to your access log and then tail the access log with a grep filter for PTST which will only show requests from WebPagetest. That way you can at least tell if it is somewhere in the application or in front of it.
It’s confirmed. It’s not the server, nor the website. Double checked with the hosting company, they’ve also tested the site from SF and there is no problem.
Obviously it’s a problem with the SF test agent that only manifests itself on the site.
The San Jose agent is now online with IE 8, 9, Chrome, Firefox 3.6 and Firefox 7. Sorry about the issues with the San Francisco agent (which I also took offline).