The blog article linked above covers it, but the short answer is that it really depends on the plugins you use and how many parameters change between variations.
For any parameter that you want to change between variations, you’ll need a widget attached to that parameter. If you’re using a B3 vst and all you’re moving is drawbar positions, then you can probably do that well with variations.
But if you’re using any of the major B3 VSTs and you’re choosing between different presets in those, there are probably a lot more parameters than just the drawbars changing. e.g., the vib/chorus mode may be different, there may be embedded effects like distortion and reverb that are different between the patches, the Leslie cabinet type and rotor speeds may be different, the capacitor models may be different (which can pretty dramatically change the sound). Unless you map more or less every parameter to widgets, changing variations isn’t going to give you the same results as changing presets in the VST itself.
For model based VSTs (which is the vast majority of what I use these days) I’ll tend to use different Rackspaces for different sounds, and my variations in those rackspaces will just be sound tweaks (e.g., chorus on/off, drawbar positions, mixer blends). If I’m using a VST that’s very RAM heavy (which tends to be load time heavy) I’ll try to use variations when I can.
On my system I’m never anywhere close to using up my RAM, so I tend to use different rackspaces more than different variations. Or sometimes a bit of a hybrid approach, where I’ll have four instances of Pianoteq in one Rackspace, and my variations will be used to turn on/off and or mix between them.
No right or wrong answer.