Asked the question on serverfault.com, but the only answer I have gotten is to switch to HAProxy, which I don't really want to do. Also the author of the response keeps suggesting keepalived, which won't work in my situation because I am not trying to establish fault tolerance for the load balancer, rather I just want the site to still be relatively performant if a pool node goes offline.
http://serverfault.com/questions/624528 … fline-down">http://serverfault.com/questions/624528/nginx-load-balancer-considerable-lag-when-upstream-node-is-offline-down
The help I have gotten from linode "support" has been disappointing.
This is essentially my last chance for a conclusive response, otherwise I may have to start researching haproxy. Which is odd because thousands of large sites use nginx as a load balancer, and must suffer node outtages from time to time. I'm sure that a 3-6 second increase would destroy some SLAs.