We revive intriguing, yet still unexplained, experimental results of Ehrlich and co-workers [Phys. Rev. Lett. 77, 1334 (1996); Phys. Rev. Lett. 67, 2509 (1991)], who have observed that 5d adatoms distributed on (111) surface islands of 5d metals favor the adsorption at the cluster's edge rather than at the cluster's interior, which lies in contrast with the behavior of 4d and 3d elements. Our state of the art ab initio calculations demonstrate that such behavior is a direct consequence of the relativity of 5d metals.