Without any PartitionHandling it is not possible to ensure the consistency of the cache.
With PH enabled it depends whether the remaining partition has the majority or at least all owners for a particular key to be able to access the key.

In some cases the prefered behaviour is to keep every entry readable but prevent from any change to keep consistency.
So there should be a configuration option for the partition handling to switch all instances to READ_ONLY instead of AVAILABLE/DEGRADED.
In this case each get will return the current value but each put will be rejected with a AvailabilityException

This prevents from any rebalancing if one partition contains a majority and stay available.
But in case of a real instance crash it might need a human interaction to set the state back to AVAILABLE as the node will never come back to the cluster.