I couldn't find such a thing either. I think I'd settle for the child having its own db handle. It won't make any difference to inherent elapsed time( because e.g. login becomes simultaneous anyway if the child does it instead), only that due to the resource impact from increasing concurrent db sessions, which itself seems inevitable anyway.
Update: I tend to use Parallel::Forkmanager for multiple DBI sessions and have the parent only manage its children rather than do any DB stuff itself during the parallel processing phase.