[PERFORMANCE] work_mem vs temp files issue

Hi,

AFAIUI, work_mem is used for some operations (sort, hash, etc) for
avoiding the use of temp files on disk...

In a client server i'm monitoring (pg 8.3.7, 32GB of ram) work_mem is
set to 8MB, however i'm seeing a lot of temp files (>30000 in 4 hours)
with small sizes (ie: 2021520 obviously lower than 8MB). so, why?
maybe we use work_mem until we find isn't enough and we send just the
difference to a temp file?

i'm not thinking in raising work_mem until i understand this well,
what's the point if we still create temp files that could fit in
work_mem...

PS: i have max_connections to 1024, i know i need a pool but the app
is still opening persistent conecctions to the db, so is not like i
could raise work_mem just easy until the app gets fixed

--
Atentamente,
Jaime Casanova
Soporte y capacitación de PostgreSQL
Asesoría y desarrollo de sistemas
Guayaquil - Ecuador
Cel. +59387171157