We have 300tb of orc data and can't convert it to Parquet. Also, ORC performs better for our use-case.
What, specifically, is preventing the conversion? I've converted hundreds of TBytes to Parquet (including moving away from that HIVE ACID stuff).
Mainly the development cost. We trust OCR and have good experience with it, it's stable, fast and compact. We don't have any experience with Parquet and it's just too hard/risky to make this kind of conversion on a production system.
parquet is much less well supported by presto. Facebook is ORC, so ORC gets implemented and optimized first