I use Parquet with ZSTD with DuckDB and it absolutely flies.
Have not reached for Spark in a long while.