Do you know how to retrieve the genes corresponding to the proportions of up-/down-regulated genes reported in the mroast() output withinthe limma package? Documentation mentions thresholds on z. However, how could you compute the mentioned z?

Well, we don't really expect you to extract the genes indicated by roast proportions. This isn't the purpose of the roast proportions. It is more sensible to identify DE genes in the usual way using the topTable() and decideTests() procedures.

If you really do want to identify the genes being counted by roast for some reason, then just use topTable() to find those genes in the set that have P < 2*pnorm(-sqrt(2)). There is no need to compute z. But be aware that this corresponds to P<0.157, which is much less stringent than hypothesis testing approaches.