The key is setting a relatively high threshold. You only want the loudest bits to be heavily compressed, leaving the quieter parts alone. That's where you get the upward compression effect, by summing those quiet parts that
aren't compressed with the same quiet parts in the main signal, making them louder.
It's still an overrated technique for many genres, though. It made more sense, for example, in the old Motown days when they were trying to cram every bit of loudness into the grooves of 45 RPM records. The fact that they were successful at that was one of the reasons for the company's amazing success (although having the likes of Smokey Robinson on staff didn't hurt, either).
But nowadays we at least have the
option of retaining dynamic range if we want to. Global parallel compression does not help achieve "punch". Not every track requires punch, e.g. vocals. But drums certainly do. So for punchy genres we're better of being selective about which elements get the parallel compression treatment.