Heap dump analysis of an OFP-centric datastore snapshot shows we are have duplicate strings retained from MapNodes - one String is retained from NodeIdentifierWithPredicates, the other one is retained from the corresponding leaf. The same is probably true for LeafSetNodes and their NodeWithValue keys.

While these duplicates are small in each instance (40-100 bytes each), there are 1.4M MapEntry nodes in the dump and strings are accounting for 18% of the allocated heap - thus we should attempt to de-duplicate them while restoring the data stream.