Single node clusters should only be deployed for testing and development work. There can be no recovery from a node failure (other than from a snapshot restore) if you have a single node. Multiple nodes will not only provide for parallel query and load operations, but will also allow for data protection, as the blocks are replicated between nodes. Additionally, as we have seen, there are some functions that will run only on the leader node. A single node cluster (the leader node) and compute nodes are one and the same, so not only do you fail to gain the benefits of parallel processing, but you actually face the penalty of not having a separate node to handle the tasks of the leader node.
When provisioning your cluster, be sure to pick a maintenance window that will work best for your load times and availability needs. It is also best, particularly early on in Redshift's life cycle, to allow automatic patching for that maintenance window. Notices are published weekly...