If a cluster of ISPN nodes was shuted down in a graceful mannor it will wait for all nodes when restarted.
Assuming that there is one or more nodes not lomger available it should be possible to enable the caches without having all nodes and might loose data if >=NumOwners are missed.

It is possible to restart the cluster by remove the 'graceful' information (unique Node ID's) from the filesystem of each node. But in this case more data is lost because the data of the first node will be rebalanced and shared for thewhole cluser after restart.