you can capture all the reject rows into a database file, since databases are designed to automatically control concurrent writes. You cannot write to a flat (sequential) file from several different process concurrently. So if in your job you point all the reject links to, say, a DataStage hashed file, and ensure that all generate a unique key then subsequently you can read that hashed file into a sequential file.
there are no hash file stages in Px as you said. You can write to UniVerse hash files from PX through ODBC (or, to be correct, by using the "Dynamic RDBMS" stage). It boils down to two generic choices - writing to several sequential files [each with 1W&NR access] and subsequently merging them into one sequential or file set OR writing to a single database table of your choice by all links. If you have many rejects then write to parallel sequential files for performance, if you have only a few rejects write to a consolidated table for simplicity. The choice is yours.