How cross-sectional heterogeneity evolves in nonequilibrium populations of competing, ageing units is a collective question that cannot be resolved from a representative trajectory or the ensemble mean alone. Analysis of more than two million age-aligned English Wikipedia page-view trajectories from six yearly cohorts spanning 2015–2020 shows that the variance of logarithmic attention grows linearly with $lnt$, while the ensemble mean relaxes on the same logarithmic clock. This dispersion law recurs across all six cohorts and is systematically favoured over the tested power-law alternative; temporal increments also exhibit consistent antipersistent scaling. A complementary trajectory-level analysis reveals broad, non-Gaussian variation in effective relaxation rates, showing that this collective regularity coexists with heterogeneous observed trajectories. A minimal scale-free stochastic model combines mean relaxation with self-affine temporal fluctuations of age-dependent amplitude. Within this model class, logarithmic spreading emerges when fluctuation accumulation balances ageing attenuation, defining the marginal boundary between algebraic broadening and saturation. Together, age-resolved cross-sectional dispersion and temporal-increment scaling provide a transferable ensemble-level criterion for distinguishing amplification, marginal spreading, and saturation in other populations of competing, ageing units.