At customer environment we see users experiencing slowness at random times at random PPS. Changes made to improve the mbuf processing. Also too many debug logs were adding to the slowness.

Tested with high PPS 70K pps IN/OUT (which more then double the PPS seen at customer place[peak time].)
below are the configs used.
log level "info"

log off
qos interface "port5" OUT 3Gb

qos enable "port5" OUT

qos interface "port5" IN 3Gb

qos enable "port5" IN

qos queue root "q_default_port5_out" "port5" OUT 3Gb 1 UNBORROW DEFAULT

qos queue root "q_default_port5_in" "port5" IN 3Gb 1 UNBORROW DEFAULT

qos sessionon

qos sespacketcnt 50

qos sessionqdelstepcount 1

qos qlenwait 1027

qos qosclasscnt 48

qos qosclspos 8