Hmmm, we've been creating big clusters in our environment without issues, but I'd suggest you try with more recent Infinispan versions, such as 5.1.5.FINAL.
Remember too that if in production, we now have a supported version of Infinispan called JBoss Enterprise Data Grid where we provide professional support:
Thanks Galder, we are planning to upgrade infinispan but will need to interact with the developers to ensure all is well.
We are using IBM Blades HS21/HS22 to host the cluster on and the infinispan is working through tomcat on these nodes. Each host on the cluster is running multiple tomcat instances (to increase the cluster size)
There has been some form of a pattern noticed with the blades, if a stop is issued to one node on a blade that has other hosts that is part of the same cluster multicast, then all the nodes on this blade chassis tend to break. Stopping starting any of these nodes will make them start their own clusters. To work around this we change the multicast IP and start up entire cluster.
Unsure if there are others who have had a similar issues with IBM blades and infinispan clustering.