The basic idea is this - on the provider, set up an accesslog DB for 
logging changes, in addition to the main database. Set up the syncprov 
overlay on the logging DB, not the main DB. Thus syncrepl will be 
propagating full entries as usual, but the content of those entries 
will actually be deltas, not actual entries... A sample config looks 
like this:
database bdb
suffix cn=accesslog
directory db.1.b
rootdn cn=accesslog
rootpw secret
...
overlay syncprov
syncprov-nopresent TRUE
database bdb
suffix dc=example,dc=com
directory db.1.a
...
overlay accesslog
logdb cn=accesslog
logops writes
logsuccess TRUE   # by default, accesslog logs failed requests too. we 
just want successful writes.
On the consumer, the configuration looks like:
database bdb
suffix dc=example,dc=com
syncrepl    rid=1
       provider=ldap://localhost:9011/
       binddn="cn=accesslog"
       bindmethod=simple
       credentials=secret
       searchbase="cn=accesslog"
       scope=sub
       type=refreshOnly
       interval=00:00:00:10
       syncdata=accesslog
This separation of log info from syncprovider is rather troublesome; 
it won't fallback properly to a full refresh if the consumer is too 
far out of date wrt the log database.
A better implementation would be to leave the syncprovider overlay on 
top of the main database, and add a config keyword telling it where it 
can find a log database to use. Then both syncprov and accesslog 
overlays would sit on the main database. And ideally, there would be a 
new flag in the syncrepl control requesting that changes be propagated 
in commit order, as deltas instead of full entries.