OK, so far I settled on excluding connection caching on app side
(Apache::DBI and prepare_cached) from equation and adding pgbouncer as a
counter-measure. This seems to stabilize the situation - at least I'm
not able to push server into high-sys-cpu stall the way how I used to do.
I'm still interested in getting to the bottom of the issue vs
alleviating it with connection poller, so I might setup a server on
Centos 6.3 (kernel ver 2.6.32) and give it a try.
Alternatively, if anyone have ideas on playing with sysctl parameters
related to scheduler - please let me know.
-- vlad