The tradeoff is sharding and caching graph DBs is difficult in terms of the performance hit because it's harder to isolate queries to a single partition/subgrap

The tradeoff is sharding and caching graph DBs is difficult in terms of the performance hit because it's harder to isolate queries to a single partition/subgraph - in some ways intersectionality is the whole point of graph vs relational. The overhead of a large scale distributed graph DB with redundancy and caching is significantly higher than a relational DB acting as a "poor man's graph DB"; and materializing your more expensive views and queries is, in the base case, a sufficient performance equalizer.