I think it's not a great habit to get into as it gives you a false sense of dynamics .. or, ...it's sorta like trying to put icing on a cake while it's still in the oven.
I see the need for craving level and cohesion, but if your playback system allows for it, I'd recommend cranking the volume knob over throwing a limiter on while mixing.
.. not saying it can't be done with decent results , but more than not , for ppl starting out, it can be a crutch and a bad habit that's hard to break. gl
I agree with what Waltz Mastering said above, but adding to it that limiting the master track can raise the power to the level that is required for more plays to take place, because of the song sounding better and have more impact at more playback levels and can make the production slightly more cohesive which can be good on various parts of the song. This however depending on the incoming dynamic balance, since a brickwall peak limiter acts on whatever transient frequencies that exceed the threshold level, so you want the peak limiter to act broadly the same across all frequencies in the transients to maintain a natural well separated sound. So limiting on the master bus is not "bad" in certain scenarios, but can add some degree of congestion if it stays the same across the whole song, that is more the issue, unless of course the peak limiter has such a character that it makes the song as a whole more pleasant and some peak limiters tend to have those kinds of very nonlinear dynamic characteristics. So there is in my view no general rules about this, other than it helps to be aware that there are cases when it can really eat on the summed signal, so you need to be aware of what you are doing when you are adding a peak limiter on the master bus, how it impacts on the production and on the sound. But be aware that a peak limiter can be the solution to a mix not having enough impact on lower playback levels. That is one of the reasons why they are used on the master bus - to lift the more quite parts of the production above the perception threshold at lower volumes, instead of the sound sources moving in and out of perception at those playback levels. Keep in mind that this can be used as an effect too, because at the same time you want the masters to grow in certain ways as you turn up the volume. For instance you might want the drums to grow in size more than some sound soruces as you turn up the volume, whereas maybe you want the vocals to stay a bit more soft on louder volumes. If you just push the mix through brickwall peak limiting on the master bus, then you might remove some of those effects without noticing it. That's why you peak limit the stems, because you want to control all of that. Peak limiting on the master bus is done mostly to lift the mix up to the minimum power level at low volume and to move the sound sources a bit closer in case the sound sources become too distant to each other, which can be an issue too.
Also want to add that the process of expanding the dynamics is as important and that it also depends on how you apply brickwall peak limiting, you can for instance apply two stage brickwall peak limiting, that might have a totally different impact on the mix. So it's essentially about knowing the characteristics of the gear you have (yes some brickwall peak limiters tend to do wonders on any content thrown at it), knowing how to best apply the limiting and to not put too much focus into it, because great mastering is also about a lot of other things. It also impacts on the mix/master setup, for instance ITB you want to use oversampling on the limiting and a heavily ITB based setup might suffer from phase issues (partly due to that), in which case you might want to focus the limiting broadly instead so that you can get less phase issues caused by the oversampling. But it's kind of not that simple, because eventhough with less input track level processing you can lower the total delay by applying fewer effects and minimize the direct phase issues between tracks, if you instead put lots of limiters with lots of oversampling on the master bus and especially since you still want to process at highest possible sample rate to reduce latency, you might end up with the same amount of total delay on the system because each input track level processing is now limited by the lower amount of available CPU to do its job. The delay is determined by the amount of processing taking place on the project as a whole, wherever that processing is, whether it is on the input tracks or on the aux/master tracks. For this very reason and especially when ITB mixes are mastered on unprinted tracks, you will quickly run into phase issues. Therefore, you want to do all you can to reduce the latency first of all, then send the signal out to the hardware domain instead and focus the heavy processing there instead. Then print straight to the final playback format hence without any needs to downsample, dither or noise shape. This is by far overall the route that yields the most beautiful sound, though it requires good hardware too. So brickwall peak limiting the master bus does not necessarily mean the same thing in the software domain as it does in the hardware domain, it all depends on the specific configurations. Now, having said that, 9 times out of 10 a hardware brickwall peak limiter sounds better than a software brickwall peak limiter. But a software brickwall peak limiter, although harsh, in conjunction with a sweet sounding hardware brickwall peak limiter can yield an even greater outcome, eventhough the software brickwall peak limiter eats more signal and hence logically appears to downgrade the sound quality as a whole. So it's really complicated. In order to really get high performance out of this you have to partly understand and control the technical aspects, but you also need to align to the multi-dimensional multi-polar nature of music, it's when you combine and balance it all, that you get high performance out of it all. Due to this complexity and the fact that this complexity has both greater positive and negative potential, early on it might be wise not to go this advanced, but instead only use your ears. Sooner or later though, you will get to the point that you will have to understand the underlying nature of it all in order to keep improving your production quality, at that point you are simply forced to open yourself to all of this complexity, there is no way around that. But at that point there are ways to simplify how you approach all of that complexity. That means the complexity is still there, but by the way you approach that it becomes easier to land at the positive potential on the scale, rather than at the negative potential of that scale. If you for instance use digital jitter to your advantage, and yes that you can do although to most that seems crazy, then it's a fine line before that becomes a disaster, the negative potential is always there and it takes a lot of understanding and work to master productions at that level. Sooner or later though, you will have to tap into the most interesting kinds of approaches, in order to be able to push your mixes/masters to the absolute best ones out there. Being a great engineer is kind of a demonstration in how well you understand sound and music and how well you approach the sound, music, the technology to make that translate into a great musical impact. After years of engineering at the very deep levels, you are going to face a situation when you need to break free from some incredibly fundamental limitations. That's maybe the hardest part of it all, but maybe also the most fun and exciting part of it... An example is going beyond the limitations of the standard equal temperament 440 Hz tuning (2^(n/12)). All of a sudden you'll be spending all of your time in Excel.