I have running a replication mechanism using SQL6.5. It concerns about 15 user tables, grouped in several categories.
The replication is running alright and steady until the transaction log on the subscriber-database is dumped with DUMP TRAN <dbname> WITH NO_LOG, which is followed by DBCC CHECKTABLE (syslogs) to correct the number of pages assigned to syslogs.
Then the replication goes nuts, only to be recovered after installing the mechanism from the beginning.
I cannot leave the DBCC out, because this seems the only way of really emptying the syslogs table.