You need to look at how the benchmark was implemented, not all are equal, but sometimes this is because of a technical problem with the library.
Relatedly, I've just found that a race condition occurs when multiple threads try to open a DB connection with db_postgres. While a lock fixes this, every dev will the same problem and need to implement a lock when they realize it's needed, so the lock really needs to be in db_postgres and possibly the other stdlib DB libraries. I filed this issue yesterday: <https://github.com/nim-lang/Nim/issues/20231>
