The best way to determine the setup that provides the best performance is to perform your own tests. For example, try loading data and runnign queries with different numbers and sizes of nodes. This way, you will have results that are accurate based upon your own particular data and that way that it is stored in Amazon Redshift.
When loading data into Redshift, the load will run better when you have more nodes because it is performed in parallel across all nodes.
In general, more nodes will always give better performance because each node adds additional storage and compute. However, you will have to trade-off performance against cost.
solved Comparison of (1, 2, 3 nodes) clusters in redshift and csv files