> Any idea of what is going on with tailfib (20% slower) or tensor (30% slower). > I assume that vector-concat (60% slower) is due to the unfold thing in the > basis library. Sad. I think you are misreading the (confusing) column headers. The new mlton is in the first column, and is faster on the three benchmarks you mention. benchmark MLton stable-MLton old-MLton