XDCR has a default checkpoint interval of 30 minutes. Replication checkpoints are written when either of 2 conditions are satisfied:
1. vBucket replicator takes longer than the checkpoint interval
2. vBucket replicator determines that it has been longer than the checkpoint interval since the last checkpoint was written
However, if you have a read-only dataset, checkpoints may never be written (initial replication finishes in less than 30 minutes, no mutations later to trigger 2nd rule).
NOTE: it should be noted that this only results in additional meta-data being sent, not full documents
In read-heavy (but not read-only) workloads the behavior may still not be optimal.