This new approach allows for the parent dataset to be the same for all the replicas and the child replicas to all send their changes to the same datasets. This is useful for a central enterprise geodatabase having the edits done by each 'field' office for their geographic district, with central serving as the enterprise data repository.
With replication many replicas can be created from the same source data but a child geodatabase hosting many replicas must have distinct datasets.
In the 9.4/10 documentation section titled Replicas and geodatabases, note 2 important statements:
1) The source must be an ArcSDE geodatabase, and therefore, parent replicas can only be hosted by ArcSDE geodatabases. You can also create multiple replicas from a single-source geodatabase. For example, you can create a replica for each county from your statewide enterprise geodatabase. The data involved in each replica may also overlap.
2) It is also possible for a single ArcSDE geodatabase to host multiple child replicas. In this case, however, the datasets involved in each child replica must be distinct. For example, if a feature class named parcels is involved in one child replica, it cannot be involved in any other child replica in that geodatabase.
