Yea, I thought parameters per compression algorithm should indeed be added in a next version :) more compute but definitely an improvement. I think pandas doesn't offer zstd as option with csv, but I'll check once more.

EDIT: indeed, it's missing in `to_csv` - seems like an oversight.