Basically it sum up that by using offset, both there the limitation mentioned of deleted records which break the offset and second that a RDBMS scan the whole table to count the index because the organisation in memory of a table is not flat