Only by incorporating various forms of feedback can theories of galaxy formation reproduce the present-day luminosity function of galaxies. It has also been argued that such feedback processes might explain the counterintuitive behaviour of `downsizing' witnessed since redshifts z ~= 1-2. To examine this question, observations spanning 0.4 < z < 1.4 from the Deep Extragalactic Evolutionary Probe (DEEP)2/Palomar survey are compared with a suite of equivalent mock observations derived from the Millennium Simulation, populated with galaxies using the GALFORM code. Although the model successfully reproduces the observed total mass function and the general trend of `downsizing', it fails to accurately reproduce the colour distribution and type-dependent mass functions at all redshifts probed. This failure is shared by other semi-analytical models which collectively appear to `over-quench' star formation in intermediate-mass systems. These mock lightcones are also a valuable tool for investigating the reliability of the observational results in terms of cosmic variance. Using variance estimates derived from the lightcones, we confirm the significance of the decline since z ~ 1 in the observed number density of massive blue galaxies which, we argue, provides the bulk of the associated growth in the red sequence. We also assess the limitations arising from cosmic variance in terms of our ability to observe mass-dependent growth since z ~ 1.