I’ve tried both methods with varying results. It all depends on the track and what you expect to accomplish with your “mastering” plugins. After all, mastering is a floating concept these days. Mastering a commercial track involves a lot of reference track a/b, adapting to radio / CD / streaming standards, and mastering an album is a lot about comparing levels and spectrums across a bunch of different tracks, not to mention track ordering, fine-tuning intros/endings etc, which doesn’t really apply to the average AJ upload.
If your stereo mixbus plugs are very heavy, like a lot of saturation, coloring EQ, or pumping compressors, then yes that would change the balance of the mix and if you slap all that on as a last “touch” you’ll be tempted to go back to the mixing stage again and again. So in that case might as well have them on from scratch.
For AJ tracks, I still tend to leave the stereo bus clean until I’m done with a lot of things on the individual tracks - including gain staging, balance, eq, dynamics, convolution etc. Then I build up my mastering chain little by little, basically trying not to change the balance and feel of the original mix too much. I may do some a/b reference eq matching, push the RMS up where it should be, smooth things out just a bit with a multiband compressor.
For me, this final stage is all about polish. It’s about listening to the mix as a whole and only addressing reference and coherence issues. To answer the OP question, if I had the mastering chain active from the get-go, which would kind of work, I would probably just add some more mastering plugins at the end anyway!
I think, from what you’re telling, that if you find yourself going back to the mix after mastering, you’re using the mastering stage with too heavy processing, heavy enough to alter the signature of the track. What you could do instead, as a middle of the road kind of thing, is add a “pre-master” stereo bus second-to-last in the chain with the heavy stuff that you obviously think you’ll need or want regardless of how your mix turns out, and keep that bus active during mixing, and then add the “polishing” mastering plugs in the Out 1-2 once you have a finished mix.
The idea behind separating these workflows, if nothing else, is to encourage creating a decent mix without chucking all your signals into a heavy stereo bus processing. In the long run this will increase your comfortability in achieving your desired soundscape without the “help” of mastering plugs, the benefits of which, IMHO, should be thought of as an “added bonus” rather than a “mixing prerequisite”.
The illusion of comfort and ease you may enjoy while blasting your tracks (possibly unknowingly poorly eq:ed and dynamically processed, at close to -0dB, etc), into a forgiving multiband compressor can, and will, eventually bite your tail, in shapes of unwanted distorsion, balance issues, pumping, offset eq and automation. Also, it’s way harder to address loudness issues in separate tracks when all audio is competing in a crowded fishbowl of compression, exciters, limiters and whatnot.
That been said, you may also have a lucky day, or grow accustomed to that kind of workflow over time and achieve astonishing results. You may feel more inspired when tracks sound “mature” from inception, which may yield better musical results, what do I know. But my take on the mixing stage is, if you depend on mastering plugs to make your tracks “happening”, you’re losing out on the opportunity to create a reliable sonic “army” before moving into mastering territory.