The way benchmarks for Composer have been presented since v1 feels unusually cautious. To users, that reads as “the model isn’t very good”.