+1 right you are about micro instances getting hammered by the hypervisor if you use too much sustained resources.
One trick that I used when I was temporarily hosting a small Clojure web app for a customer: I ran the web app using "nice" to reduce its priority. I did not do any measurements, so this is really subjective, but the app seemed to run a lot better as far as consistently getting a little bit of processor time to run.
Not so subjective: I did some Clojure development in a repl on a micro instance (because it was already set up for access to a Hbase cluster) and doing a "nice lein repl" really made development possible, if a little slow.
One trick that I used when I was temporarily hosting a small Clojure web app for a customer: I ran the web app using "nice" to reduce its priority. I did not do any measurements, so this is really subjective, but the app seemed to run a lot better as far as consistently getting a little bit of processor time to run.
Not so subjective: I did some Clojure development in a repl on a micro instance (because it was already set up for access to a Hbase cluster) and doing a "nice lein repl" really made development possible, if a little slow.