As part of an effort to benchmark open source SPARQL engines on Wikidata, the page https://www.wikidata.org/wiki/Wikidata:Scaling_Wikidata/Benchmarking/Existin... contains some initial results and analyses of benchmarking Blazegraph, MilleniumDB, QLever, and Virtuoso on several existing SPARQL query benchmarks for Wikidata. There are some surprising results there, particularly related to different answers produced by different engines. Suggestions on how to improve the effort or provide deeper explanations of the results are particularly welcome on the discussion page.
Peter F. Patel-Schneider