If you are going to query it frequently then json/csv might become an issue. I think the reason it doesn't become a problem for duckdb/polars users is that we generally convert them to parquet after first read.