How do I reduce database corruption?

[Dan Shriver <dshriver@sharemedia.com>]

Hello,
    I have the following problem: about once a week a user is
editing my LDAP database and they corrupt it.  Furthermore
whenever this happens slapindex cannot fix the problem, I have
to slapcat it to a file, remove the createTimestamp...
attributes, ldapmodify -a it back into the directory (after it
was emptied) multiple times as the ldif is in no particular
order.
   I have not seen this problem arise from ldap tools
(ldapmodify, ldapdelete...) I have only seen it result from
ldapbrowsers such as: GQ, ldapBrowser, ldapViewer- but then the
users rarely use the tools.
  I am now running 2.0.21 and I have removed the increase speed
/ decrease stability slapd.conf options such as
dbcachenosynch..., so I would assume that I should have a stable
system.
   Also I find the whole thing disturbing as the browsers should
be using the standard LDAP protocol (so how could they screw
things up?).
   In general the corruption results from a bad modification
request such as:

   dn: cn=x,o=y
   cn: x

       to

   dn:cn=z,o=y
   cn: x

       but if this screws up the indicies shouldn't the protocol
send a blocking message back to the browser, and halt the
operation?
   There are a number of people on this list using openLDAP for
big sites with lots of hits (so far my hits are pretty low) so
obviously I'm missing something.  Any tips?