Work was completed recently that successfully demonstrated the value of implementing threading affinity:

• http://www.google-melange.com/gci/task/view/google/gci2012/7960222

This task involves cleaning up that work and polishing it into production-quality code.  That nominally involves the following:

• hooking behind bu_parallel(), not the individual worker functions
• unwrapping the new function so it's always available
• documenting the function properly (missed that mistake before closing the task)
• pulling the static globals into the callback in parallel.c
• creating a separate callback that takes an arg
• make the one with static data call the new arg one
• updating the callback to utilize the arg parameter, pass it during pthread_create()
• update the other platforms to do the same where we can, and finally
• sorting out what to do with _GNU_SOURCE (since just testing means it'll never be set)

Submit your work as a single patch file.  Note this task requires meticulous attention to detail.