Computational strategies for the treatment of relativistic effects including spin-orbit coupling at a highly correlated level are compared for a number of heavy atoms: indium, iodine, thallium, and astatine. Initial tests with perturbation theory emphasize the importance of high-energy singly excited configurations which possess large spin-orbit matrix elements with the ground state. A contracted basis consisting of L-S CI eigenfunctions (LSC-SO-CI) is found to give an accurate representation of both spin-perturbed P-2(o) components as long as key np-->p(i)* singly excited configurations are included. Comparison is made with a more extensive treatment in which all selected configurations of various L-S symmetries form the basis for the multireference-spin-orbit-configuration interaction (MR-SO-CI). Good agreement is obtained with experimental SO splittings for the In, I, and At atoms at a variety of levels of treatment, indicating that the L-S contracted SO-CI approach can be implemented quite effectively with relativistic effective core potentials (RECPs) for both very electronegative atoms and also for lighter electropositive elements up through the fifth row of the periodic table. The thallium atom SO splitting is more difficult to obtain accurately because of greater differences between its valence p(1/2) and p(3/2) spinors than in the other cases studied, but good results are also possible with the contracted SO-CI approach in this instance, provided proper care is given to the inclusion of key singly excited L-S states. The relationship between all-electron two-component SO-CI treatments and those employing RECPs is also analyzed, and it is concluded that triply excited configurations relative to the P-2(o) ground state are far less important than previously reported.