Just some corrections.
First, you don't need to run a separate cluster for spark. You have YARN.
And with MR2 (YARN) you could in theory run storm on the same cluster. Albeit you have to monitor the cluster and storm which will run outside of YARN to make sure that you're not overloading the cluster machines.
It will be interesting to see more on Streams. To date, MapR has built better mousetraps when compared to Apache's release(s).