"In the example provided for requiring a restart (some Transport properties), this may require a restart of all nodes in the cluster. How do you model this?"
If the cluster is a collection of standalone instances, there is no cluster-wide management support, so making a change on one host will not be propagated to other hosts in the cluster. Each cluster node needs to be separately changed and separately restarted (in this case) by the administrator.
If the cluster is a server group in a managed domain, changes (made via the management console) to the Infinispan subsystem in the server group's profile will be propagated automatically to all started hosts in the server group, putting all servers in the "process-state => restart-required" state. The servers in the server group then need to be restarted, which can again be done using the management console.