I upgraded one of my Jenkins instance plugin to 1.26. I still see the zombie issue (but probably not as severe). From Chris' comment about the master lost connection to slave, I am not sure if that is my original case which I have instances with all the tagging key-value set but not the "name" tag.
If the slave has ever connected to the master once, and some how lost the registration, I do not think the behavior observed would fall into this situation.
I do have another ticket about "out-of-sync" issue
JENKINS-26798 - High occurrence of out-sync EC2 slaves