That was probably about what I got when I migrated some heavy number crunching code from Pandas to Polars a few years ago. Maybe even better than that.