If you increase the number of replicas, then you have to rebalance to create them.
If you manually increase the number of replicas of a bucket, the operator will return it to its original value.
However the logic for rebalance is naive and just checks if one is required, so if you increase replicas then the operator will rebalance.
There should probably be a quick check to see if the value for the replicas of each bucket is the same as the one specified in the cluster definition.
In that case the operator can revert it to its original value and should do so without incurring a rebalance.