We present a comprehensive study of the observational dependence of the mass-loss rate in stationary stellar winds of hot massive stars on the metal content of their atmospheres. The metal content of stars in the Magellanic Clouds is discussed, and a critical assessment is given of state-of-the-art mass-loss determinations of OB stars in these two satellite systems and the Milky-Way. Assuming a power-law dependence of mass loss on metal content, dot{M} ∝ Zm, and adopting a theoretical relation between the terminal flow velocity and metal content, v_∞ ∝ Z0.13 (Leitherer et al. 1992, ApJ, 401, 596), we find m = 0.83 ± 0.16 for non-clumped outflows from an analysis of the wind momentum luminosity relation (WLR) for stars more luminous than 105.2 {L&sun;}. Within the errors, this result is in agreement with the prediction m = 0.69 ± 0.10 by Vink et al. (2001, A&A, 369, 574). Absolute empirical values for the mass loss, based on Hα and ultraviolet (UV) wind lines, are found to be a factor of two higher than predictions in this high luminosity regime. If this difference is attributed to inhomogeneities in the wind, and this clumping does not impact the predictions, this would imply that luminous O and early-B stars have clumping factors in their Hα and UV line forming regions of about a factor of four. For lower luminosity stars, the winds are so weak that their strengths can generally no longer be derived from optical spectral lines (essentially Hα) and one must currently rely on the analysis of UV lines. We confirm that in this low-luminosity domain the observed Galactic WLR is found to be much steeper than expected from theory (although the specific sample is rather small), leading to a discrepancy between UV mass-loss rates and the predictions by a factor 100 at luminosities of L 104.75 {L&sun;}, the origin of which is unknown. We emphasize that even if the current mass-loss rates of hot luminous stars would turn out to be overestimated as a result of wind clumping, but the degree of clumping would be rather independent of metallicity, the scalings derived in this study are expected to remain correct. Appendix A is only available in electronic form at http://www.aanda.org