A method for improving replication persistence in a caching appliance structure can begin when a primary catalog service receives a command to instantiate a data partition. The primary catalog service can manage a collective of caching appliances in a networked computing environment. The data partition can include a primary shard and at least one replica shard. The primary shard of the data partition can be stored within a memory space of a first caching appliance. The at least one replica shard of the data partition can be stored within a non-volatile storage space of a second caching appliance. The first and the second caching appliances can be separate physical devices. The memory space of the second caching appliance that could have been used to store the at least one replica shard can be available for storing primary shards for other data partitions, increasing the capacity of the collective.