If you are doing individual DML commands (insert, update, delete), you'll be fine for a very long time.

Where the flat file begins to have problems is if you have 100 inserts to do, and the flat-file interface (e.g., DBD::CSV) wants to read and write the file for each command.

Why not test for your specific case? Create a large dataset and see how your app performs.