Hi,
An issue is how to configure the network of master-ami and compute-ami instance nodes given EC2's current network.
The following project seems to have tackled the problem outside of an Globus/Nimbus software stack:
"ElasticWulf: Beowulf cluster run on Amazon EC2"
http://code.google.com/p/elasticwulf/
Discussed initially here:
http://www.datawrangling.com/mpi-cluster-with-python-and-amazon-ec2-part...
HTH?
Mark