I think you could compute the hash value for the join columns, then join the two dataframes using that hash value. It will save the cost to match the join conditions.

Polars provides a built-in hash function: https://docs.pola.rs/api/python/stable/reference/expressions/api/polars.Expr.hash.html#polars-expr-hash, or try other hash functions provided by https://github.com/ion-elgreco/polars-hash