Lithospheric strength profiles generated for a shortening continental lithosphere generally predict excessively high differential stresses in the sub-Moho continental mantle; this seems inconsistent with the relative scarcity of earthquakes at this depth. This inconsistency was put forward as evidence for weak mantle rheology. However, this argument implicitly assumes that strength envelopes are valid in actively deforming regions. We test this assumption on two end-member model lithospheres having identical upper crustal rheologies, but with (1) a weak lower crust and strong mantle, and (2) a strong lower crust and weak mantle. For this purpose, we compare one-dimensional (1-D) with 2-D visco-elastoplastic numerical models of continental shortening. The 2-D models show that strongly heterogeneous deformation typically follows initially homogeneous deformation. Lithospheric-scale buckle folds and shear zones result in strain rate variations of as much as three orders of magnitude. Differential stresses in the upper crust are close to yield, as predicted by 1-D models. Stresses in deeper lithospheric regions, however, are significantly smaller than in 1-D models, especially in actively deforming regions. Systematic numerical simulations as a function of temperature and deformation rate reveal that 1-D models are reliable in hot and/or slowly deforming lithospheres only. The relative scarcity of earthquakes at mantle levels should thus be interpreted as an intrinsic consequence of strong lithospheric deformation rather than as evidence for a weak upper mantle.