Note: Since I did this, there have been a handful of releases of nss_ldap, and two releases of openldap, not to mention my serving tuning skills improving, so these numbers may be even less accurate now.
create 5000 some odd files in /tmp/ldap_test chown them so that every id currently in NIS gets about 12 files each
The approximate results:
Increased the ldap caches, bumped the log level down, and in theory "indexed" some common data:
those are average numbers, and all test were after doing enough to assume those files are cache'd approximately the same.
This is with an ldap server thats a 2xpIII 500/512m ram, and nis just the local nis server here (a p100 with 256m ram).