I do not think there is any magic vacuum thickness such as 15 Ang to assume that artifacts of using periodic boundary conditions are eliminated. One should always verify if the results are converged with respect to this parameter.
Note that the best way to avoid this issue is to use Dirichlet/Neumann boundary conditions in left/right vacuum region for slab calculations, and Dirichlet and Neumann for the left electrode and right vacuum region for surface Green's function calculations, i.e., for one-probe device. One still has to choose vacuum regions carefully, checking that there exist sufficiently large, flat potential (e.g., HartreeDifferencePotential) regions in vacuum near the boundaries.