Data fragmentation- fixes?

["matthew sporleder" <msporleder@gmail.com>]

I have a unique situation relating to my conversion from 2.1 to 2.3.
What I did:
slapcat -c -f 2.1.slapd.conf
remove entryCSN
slapadd -c -q -f 2.3.slapd.conf

This worked well, but I got a few errors about duplicate data.  It
turns out that my 2.1 database  had multiple entries of some
near-the-top data.
ex:
dn: dc=com
dc: com
objectClass: dcObject
objectClass: organization
o: Example Org
structuralObjectClass: organization
entryUUID: 0ac313d0-4aba-1099-9dd8-f266f9337e15
creatorsName: cn=Manager,dc=example,dc=com
modifiersName: cn=Manager,dc=example,dc=com
createTimestamp: 20030720064923Z
modifyTimestamp: 20030720064923Z

<100 lines later>
dn: ou=SITE1,ou=people,dc=example,dc=com
ou: SITE1
objectClass: top
objectClass: organizationalunit
structuralObjectClass: organizationalunit
entryUUID: 80a3eee6-2c61-1028-8468-973a8421431b
creatorsName: cn=Manager,dc=example,dc=com
createTimestamp: 20040427064022Z
modifiersName: cn=Manager,dc=example,dc=com
modifyTimestamp: 20040427064022Z

<millions of lines later>

dn: ou=SITE1,ou=people,dc=example,dc=com
ou: SITE1
objectClass: top
objectClass: organizationalunit
structuralObjectClass: organizationalunit
entryUUID: c41ed3f6-2c62-1028-8418-82cea5fe7b2c
creatorsName: cn=Manager,dc=example,dc=com
createTimestamp: 20040427064925Z
modifiersName: cn=Manager,dc=example,dc=com
modifyTimestamp: 20040427064925Z

dn: cn=someuser,ou=SITE1,ou=people,dc=exmaple,dc=com
etc: bar

dn: cn=someotheruser,ou=SITE1,ou=people,dc=exmaple,dc=com
etc: foo


Now, during my slapadd, the FIRST definition of SITE1 was added, but
it's really far away from all of its data. (the second one gave an
error, so at least we have more correct data)  The rest of my SITEn's
are really close to their data, so searches based in
ou=SITEn,ou=people,dc=exmaple,dc=com for objectclass=* return very
quickly with top-level info and everything under it.  For SITE1,
however, there's a -long- gap.

I tried to reindex, and now I can find the top-level definition
quickly, but the rest of the data is still slow to find (it appears to
be searching for everything).  Is there a way to get the data
re-ordered so it's closer to where its children without slapcat,
reorder manually, and slapadd?  (would that even work?)

(By the way, this only hurts searches like the one described above,
looking for actual people works the same throughout the dataset.)