It has recently been shown that galaxy formation models within the LambdaCDM cosmology predict that, compared to the observed population, small galaxies (with stellar masses < 10^{11} M_sun) form too early, are too passive since z ~ 3 and host too old stellar populations at z=0. We then expect an overproduction of small galaxies at z > 4 that should be visible as an excess of faint Lyman-break galaxies. To check whether this excess is present, we use the MORGANA galaxy formation model and GRASIL spectro-photometric + radiative transfer code to generate mock catalogues of deep fields observed with HST-ACS. We add observational noise and the effect of Lyman-alpha emission, and perform color-color selections to identify Lyman-break galaxies. The resulting mock candidates have plausible properties that closely resemble those of observed galaxies. We are able to reproduce the evolution of the bright tail of the luminosity function of Lyman-break galaxies (with a possible underestimate of the number of the brightest i-dropouts), but uncertainties and degeneracies in dust absorption parameters do not allow to give strong constraints to the model. Besides, our model shows a clear excess with respect to observations of faint Lyman-break galaxies, especially of z_{850} ~ 27 V-dropouts at z ~ 5. We quantify the properties of these "excess" galaxies and discuss the implications: these galaxies are hosted in dark matter halos with circular velocities in excess of 100 km s^{-1}, and their suppression may require a deep re-thinking of stellar feedback processes taking place in galaxy formation. Comment: 17 pages, 13 figures, 1 table; accepted for publication by MNRAS