Databases are just much, much faster than Pandas, and that's before you start factoring the extraction and loading of data. I treat Pandas as a last resort when I can't do something in SQL, generally this is something like integrating with external services or running recordlinkage.