With any Distributed Data system - you have to assess how you are intending to support failover
When Objects/Data are stored in a JavaSpace it is up to the underlying server/service application as to how that item is persistently managed
Since 'Clients' hunt for the JavaSpace service you can design the system such that many 'servers' offer the service - when one goes down the 'Client' just hunts for another in the network.
The replication of Objects/Data across these server applications is something you will have to design into your system - so that a 'Client' attached to one JavaSpace service can manipulate an Object/Data item & doesn't care that the originator of the item has connected through a different server.
Obviously your network must include several replicated Service lookup registries