mhh.. seems that - when managed by pacemaker/pcs - when resource's op timeouts are not long enough, then squid might do such "coredump"
Tuning those fixes the "issue", so it seems.
thanks, L.