C3P0 was slighlty less efficient, but the main reason to switch is that I was having an hard time to get reproducible results: with Hikari I can actually run the same benchmark twice and get close enough results, so it's giving me a better confidence in the data.
And I wanted to try it