So after waiting 24 hours and then looking at the AWS billing reports a couple of things stand out. At first we were puzzled as to why we were billed for Run Instances as well as Spot Instances. Turns out that when you launch a cluster the head node will be provisioned identically to the worker nodes unless you specify a different image type. The config entries below will create a cluster with 5 worker nodes and 1 head node.
CLUSTER_SIZE = 6
NODE_IMAGE_ID = ami-999d49f0
NODE_INSTANCE_TYPE = m2.2xlarge
MASTER_INSTANCE_TYPE = m1.small
MASTER_IMAGE_ID = ami-899d49e0
The worker nodes will be 64 bit 13xEC2 servers and the head node will be a 32 bit 1xEC2 server. This means you won't be billed disproportionately for a very powerful head node when a smaller instance will do. Bear in mind that the head node is not a Spot Instance only the worker nodes are so you need to factor this into your cost estimate. Also remember that for large or very busy clusters a more powerful head node may be required.
As an aside, yesterday's test came to $3.67 (about 29 ZAR).
The table above indicates the rather large cost saving one can achieve by making use of Spot Instances. This is based on our test yesterday. The Run Instance column is what one would pay for a guaranteed instance.