Small correction - that table 1.5 vs. 12 was from some other test. This is what I meant to post (hfsc, tbrsize 12k, bandwidth/realtime/link/upper set to the same value, although anything besides upperlimit didn't matter much):
theoretical / actual 8 / 8 10 / 10 12 / 11.95 14 / 13.24 16 / 14.13 18 / 14.18 20 / 14.27 22 / 14.16 24 / 14.50 26 / 14.70 28 / 14.60 30 / 16.36 32 / 16.48 34 / 16.38 36 / 16.39 38 / 16.38 40 / 17.28 42 / 18.88 44 / 19.46 46 / 24.82 48 / ~25.5 50 / ~27