Did you run the mysqltuner and do any of the other tuning recommendations like adjusting polling threads. When I had a single instance, I was hitting 500 seconds for polling. After using the mysql tuner script and doing the recommended changes, I dropped my poller time in half.
You can also do two full clones and implement a three node galera cluster with redis and rrdcached to get a little fault tolerance.
Since I transitioned to a 5 node poller setup, I have doubled the amount of devices I am polling but can now poll around 2k devices with the longest poller taking 58 seconds to complete.