you could add a flag item to the table, and in addition to printing 'different' you would update the flag to an appropriate value (say, 1). You would need to use an Update Cursor instead of a Search Cursor, of course. After setting the flag, just export the rows with the flag = 1
Then drop the flag item
OR
keep the search cursor, but when you find a 'different' row, capture all the attributes (including the Shape) for that record to a python list.
append that list to a master list, so you build a list of lists of records.
you will also build a similar list of the table items, types, and widths, appended to a master list for the table
Create a new feature class, with the same table schema (use the table items list to add items)
feed the master records list to an Insert Cursor, breaking out each item value (and the shape itself) from the nested list.