DAG serialization stores a serialized representation of DAGs in the metadata database.
With 1000 DAG files, the webserver shouldn't re-parse every Python file just to draw the UI. Serialized DAGs let components read the structure without that cost.
You get less parse overhead, a webserver that scales better, and a cleaner split between "parse DAGs" and "serve the UI."