It seems that the problem was a combination of 2 missing settings. The first forces the JRE to prefer ipv4 and not v6. This was necessary (I guess) since we tend to try to connect to that via a v4 address:
The real blocker was the very fact that JMX works by initially contacting the RMI port that responds with the hostname and port for the JMX client to attach. With no further settings, it'll use the local IP of the box that is a 10.X.X.X virtual address which a remote client cannot route to. We needed to add the following setting that is the external hostname or IP of the server -- during this case, it's the elastic hostname of the server.
The trick, if you're attempting to automate your EC2 instances (and why the hell would you not), is the way to find this address at runtime. To do that you just need to put something just like the following in our application boot script:
# get our _external_ hostname
RMI_HOST=`wget -q -O - http://169.254.169.254/latest/meta-data/public-hostname`
java -server \
-Djava.net.preferIPv4Stack=true -Djava.rmi.server.hostname=$RMI_HOST \
-jar foo.jar other parameters here > java.log 2>&1
The mysterious 169.254.169.254 IP within the wget command above provides information that the EC2 instance will request about itself. I'm disappointed that this doesn't include tags that are solely available in an authenticated call.
I initially was using the extern ipv4 address but it looks like the JDK tries to make a connection to the server-port when it starts up. If it uses the external IP then this was slowing our application boot time until that timed out. The public-hostname resolves locally to the 10-net address and to the public-ipv4 externally. So the application now's beginning fast and JMX clients still work.