Keeping the node metadata and their values in separate column families may be a significant performance boost. It would require much less data to be read and written when updating intermediate nodes where disk/memory throughput is the main bottleneck.