For VCA, as the construction is as simple as averaging the potentials of atomic species, the main advantage is computational. This is in the sense that it is very simple to change the compositional ratio
of a solid solution by just changing how you average the potentials, instead of having to work with
supercells (as you would have to do with SQS or other approaches even for simple ratios).
However, simply by doing this, you can imagine that the new potential will not be physical and neither the
calculations associated to it. To give some examples, properties which depend on the local
environment won't be correctly reproduced (e.g., the band gap in GaAlAsN and GaPN and GaAsN).
Then, how is it that VCA has been shown to be successful for same cases? Well, it happens that for some materials, the contributions of some atomic species are very similar to each other. Thus, averaging their potentials doesn't change the general properties (so much). Having said this, the real limitation of VCA is that it depends on how similar the parent atomic species are to each other.