Linear embedding via Green's operators (LEGO) is a domain decomposition method for solving electromagnetic problems which involve composite 3-D structures. By combining the standard Method of Moments with a set of macro basis functions obtained through the Arnoldi iteration, the algebraic system of LEGO can be effectively compressed. However, under general circumstances it is not easy to choose the minimum number of macro basis functions for a given level of accuracy, since many physical and geometrical factors come into play. To shed light on this topic, we discuss a few examples with a focus on the convergence of the numerical solution