We investigate the effect of different star formation histories (SFHs) on the relation between stellar mass (M-*) and star formation rate (SFR) using a sample of galaxies with reliable spectroscopic redshift z(spec) > 2 drawn from the VIMOS Ultra-Deep Survey (VUDS). We produce an extensive database of dusty model galaxies, calculated starting from a new library of single stellar population (SSPs) models, weighted by a set of 28 different star formation histories based on the Schmidt function, and characterized by different ratios of the gas infall timescale tau(infall) to the star formation efficiency nu. Dust extinction and re-emission were treated by means of the radiative transfer calculation. The spectral energy distribution (SED) fitting technique was performed by using GOSSIP+, a tool able to combine both photometric and spectroscopic information to extract the best value of the physical quantities of interest, and to consider the intergalactic medium (IGM) attenuation as a free parameter. We find that the main contribution to the scatter observed in the SFR-M-* plane is the possibility of choosing between different families of SFHs in the SED fitting procedure, while the redshift range plays a minor role. The majority of the galaxies, at all cosmic times, are best fit by models with SFHs characterized by a high tau(infall)/nu ratio. We discuss the reliability of a low percentage of dusty and highly star-forming galaxies in the context of their detection in the far infrared (FIR).