I'm not sure I understand your concern. The same model could be tested as a single step/simultaneous model using path analysis with both dependent variables (B and C) and all independent variables included in a single model. You would obtain the same coefficients. So I do not understand why you would be concerned about "two different statistical tests." The approach that you describe merely breaks down the complete model with two outcome variables (B and C) into two sub-models. Breaking it down is not necessary either conceptually or statistically. Both equations could be considered simultaneously using path analysis. The reason for breaking down the model is only technical in nature so that the parameters can be estimated using single-outcome multiple regression analysis.

Dear Christian Geiser , thanks for your answer and apologies if I was a bit cryptic in my description.

My concern is that, if talking about a model and depicting it in a paper with a single figure, it should be necessary to test the model with all the variables included. As you said with a path analysis, for example.

If testing two distinct sub-models, these can maybe confirm two different hypotheses, but they do not support a whole comprehensive model and do not allow the depict a unique Figure.

I do not know if you agree with this..?

It certainly has advantages to estimate the parameters of a model with multiple outcome variables as a single-step (simultaneous equation) model (path analysis). I would always recommend doing so for practical reasons, and also so one can study a test of overall model fit for overidentified (df > 0; non-saturated) models (i.e., chi-square test of model fit). Such a test is not available within a multiple regression framework.

However, for saturated (just identified) mediation models (df = 0), it really does not matter whether you estimate them in a single step (path analysis) or multiple steps (multiple regression analysis). You should get the exact same direct effects (regression/path coefficients), and therefore also the same indirect (mediated) and total effects. Therefore, I would not generally agree with you that the parameters of a single model with multiple outcome variables could not be estimated in several steps/analyses.