The version is 2.4.23, the config is the same across all the slaves, including the ones that stay in sync under load, the hardware itself is more than capable, I'm curious however to your clock comment, would an unstable clock on the slave cause the contextCSN token to be out of whack? Hence me asking rather how the contextCSN is generated and passed from master to slave, rather than, here's all my configs fix it for me.
It is required that the clock be in sync on all servers. This is well documented. You may also wish to read <http://www.openldap.org/lists/openldap-technical/201109/msg00047.html>
I would also note your version of OpenLDAP is extremely old, and has numerous known issues with sync replication that have since been fixed. See <https://www.openldap.org/software/release/changes.html> for more details.
--Quanah -- Quanah Gibson-Mount Sr. Member of Technical Staff Zimbra, Inc A Division of VMware, Inc. -------------------- Zimbra :: the leader in open source messaging and collaboration