It depends on the use case. I’m trying to build a mixer based on the Duo X and an USB audio interface. I experimented with two different approaches: First, dedicated volume plugins (almost 40 instances) for each bus. Second, cascaded mixers (11 instances) for each bus.
In both scenarios CPU usage was pretty much the same while I even got VU-meters, pan and mute included in the mixers. The only disadvantage is that the usage of the AUX bus in the mixer plugin is muting the signal on main which complicates stuff like send effects. Working on my own fork for that.