it depends on what you do, and how many columns data have.
Batch Job in data of many columns -> Parquet, Polars is best. it can read few of specific columns you need. it cnsumes less memory. So, faster.
Batch Job in data of few columns -> choose the more comfortable for you. But, if your machine's performance(CPU, memory) is fairly lower than server, I think database would be faster.
Batch Job in data of Large columns -> Server(Database) would be faster. But, if the server already stored data in parquet, Polars would be faster.
ACID Job with single transaction -> only Database