It's true that the FreeBSD process is far more likely to result in an ideal design, but Linux's approach of "build it fast, replace it fast" might average as a win in terms of total number of tolerably performing features per development time, and volume of the search space explored.