The high end of the stellar mass function of galaxies is observed to have little evolution since z~1. This represents a stringent constraint for merger-based models, aimed at explaining the evolution of the most massive galaxies in the concordance ΛCDM cosmology. In this Letter we show that it is possible to remove the tension between the above observations and model predictions by allowing a fraction of stars to be scattered to the diffuse stellar component (DSC) of galaxy clusters at each galaxy merger, as recently suggested by the analysis of N-body hydrodynamical simulations. To this purpose, we use the MORGANA model of galaxy formation in a minimal version, in which gas cooling and star formation are switched off after z=1. In this way, any predicted evolution of the galaxy stellar mass function is purely driven by mergers. We show that, even in this extreme case, the predicted degree of evolution of the high end of the stellar mass function is larger than that suggested by data. Instead, the assumption that a significant fraction, ~30%, of stars are scattered in the DSC at each merger event leads to a significant suppression of the predicted evolution, in better agreement with observational constraints, while providing a total amount of DSC in clusters, which is consistent with recent observational determinations.
