that's exactly right. like my library doc says,
# This allows us to easily make our locking as fine-grained as we want
# -- each set/key combination identifies a unique mutex.
# Mutexing becomes useful whenever you want to make sure different
# interpreters aren't steping on each others' toes. E.g. for
# on-demand ("lazy") data loading, like the original util_memoize.
# Or when your script follows the "if A, then B" pattern, where executing B
# potentially makes A no longer true, and you want to make sure that multiple
# requests for the script can't cause B to execute many times by different
# threads grabbing the same value of A before any of them execute B.
For carnageblender almost all the places I use ad_with_mutex would
cripple throughput if I had to use a Big Hammer of a single global mutex. Although I sometimes do use a Big Hammer; that is easier and more readable with ad_with_mutex than with rolling them manually, as well.
