Comment by aeonsky

Yes, my team had a direct issue with this on Aurora Postgres, at least. This is PG9 but then kept happening all the way into PG12 until we got rid of all but like 5 roles. Above like 4000 roles we experienced a significant lag on every query, sometimes on the order of seconds. At scaled somewhat linearly. I even wrote to Tom Lane and he said that area of Postgres is poorly optimized.