Correct way to handle multitimbral VSTs?

Most people do not use instrument tracks for multi-timbral synths because they do not offer as much versatility as MIDI-Audio track pairs. But they'll work if that's what you're more comfortable with, just with fewer options.
 
You're usually going to set up the sampler one of two ways: separate outputs for each instrument routed to separate audio tracks, OR use one stereo pair of outputs into a single stereo track and do all the mixing within the sampler itself. The former method is best for dissimilar instruments, e.g. a percussion instrument + a piano + a string section, or for a mix of mono and stereo instruments. For like instruments, e.g. string ensembles with one instrument each for violins, violas, cellos and basses, it's more convenient to mix them within the sampler and then treat them all as one instrument for the main mix.
 
Usually, each instrument will get its own MIDI channel, so if you stack 16 instruments in one sampler instance you'll necessarily have 16 MIDI tracks. Sometimes two instrument might share a MIDI channel, such as when layering kicks, generic string patches or synth leads. Sometimes you'll want to clone a linked MIDI track to drive two instruments from the same MIDI data (perhaps transposed). "Instrument Tracks" don't give you these options.
 
I do not have a single method, and don't believe there is any one "proper" way to do it. It comes down to whatever works for you. I will, however, note that more often than not, I'll use a separate MIDI/audio track pair for each instrument; 16 instruments = 32 tracks. I like to keep all my options open.