After their formation, mountainous landscapes gradually evolve toward smoother geometries controlled by the interplay of erosion and sedimentation. The statistical mechanical properties of this process and the link between topography and geology have remained largely unexplored. We analyze the slope statistics of different mountains worldwide, showing that landscape age is fingerprinted in their distribution tails. Data reveal a universal relaxation process, through an algebraic decay progressively replaced by an exponential one, with exponents described by a global monotonic function. We then investigate the dominant components of this dynamic smoothing using a landscape evolution model, showing that the time evolution of slope statistics results from a delicate balance between diffusive soil creep, noise, and advective river incision, with the relaxation phase mainly dominated by diffusion. Results may suggest ways to formulate reduced order topographic evolution models for geomorphological and climatological applications, and to explore similarities in surface evolution in different contexts.