You could use a checksum to catch all the data that is the same. Nice and easy, you could use the checksum as a primary key, or a hash key.

Not sure about small changes. You obviously have a method of retrieving the data, so perhaps that method could look for similarities or patterns, then only retrieve the delta.

Its kinda had to give any concrete answer with out any specifics. If are able to provide an example of the data model it would make it easier to answer.