I've re-tested with git RE24 65a49595f563573f91da936ea27f5eaf6b21a48d with *nearly* the same data (data2) and the bug does not seem to happen again. Hmm...
But with the old last good backup LDIF (data1) the bug happens again.
I noticed that in data1 the entries which were mixed up later in slapcat output were in the order (example data)
dn: cn=John Doe,o=Company
dn: o=Company
whereas later in data2 there was exported:
dn: o=Company
dn: cn=John Doe,o=Company
I will try to provide a test LDIF if necessary but I'm in a hurry now.
Please note that I'm talking about LDIF files which were all generated by slapcat and therefore are assumed to be valid input for restoring with slapadd.
Also LDAP access to the entries mixed in slapcat output is completely ok.
Ciao, Michael.