One option you could consider if you have the transaction logs from the subscriber is a detailed analysis of these using a third party log reader or the undocumented function fn_dump_dblog. Check Paul Randals blog for details on this function and the warning on its usage. What you describe just should not happen but the transactions logs should give you the options for a forensic investigation and you should be able to show what happened all the way from the publisher through to the subscriber.

While the table size may have precluded this but did you consider tablediff which could have generated the tsql to sync the table at the subscriber which may have been quicker than re-initialising