Comment by tpetry

Most parallel operations in PG are implemented by simple merge the dataset, work independently and merge the results. I expect the new distinct to behave the same and not work on a shared data structure.