through my testing, I confirmed Tom's guess - performance benefit of going beyond the first level keys is negligible. Getting top level key's JSON part of the document and use ->>, -> operators for extracting data works very well, even if I had to go down multiple levels of hierarchy. Thanks a lot!