Hmm seems like improving wikidata’s API performance would be a good use of Wikimedia’s war chest.
Their main blocker is lack of a high-performance SPARQL backend, for complex queries where ordinary text-based search is not effective. https://www.wikidata.org/wiki/Wikidata:SPARQL_query_service/... has more information, though it's not fully up to date (Qlever especially has had significant updates since).
What was dissapointing last time I used it is arguably not a problem of openrefine at all: the connectivity and response of wikidata queries was very slow. But that combination of local data harmomization with an open and globally available reference is super-important. I hope it somehow receives more attention and traction.