Unlike M and C stars, S stars are widely scattered in the (-,-) diagram. They also exhibit various spectral features attributed to silicates, SiC or a combination of both. Therefore, they do not fit easily in the ''interrupted mass loss'' scenario proposed by Willems & de Jong to account for the respective locations of M stars, optical and IR carbon stars in that diagram. A first step towards a better understanding of the evolutionary status of S stars has been achieved by the identification of two different kinds of S stars: binary, Tc-poor S stars have no or very moderate IR excesses and are the descendants of barium stars, whereas Tc-rich S stars are genuine AGB stars, probably representing transition objects between M and C stars. Still, the IR excesses exhibited by Tc-rich S stars cover a wide range, from moderate to very large at 12, 25 and/or 60 mum. The present program aims at testing the hypothesis that luminosity is the hidden parameter responsible for this variety of IR excesses, as expected if mass loss increases with luminosity along the AGB. Observations of late-type giants in clusters indicate that a good luminosity indicator is provided by the width of the lines, which is dominated by macroturbulence. That indicator has been obtained for a large sample of S stars observed with the spectrovelocimeter CORAVEL, and has been calibrated on the giant branch of 47 Tuc. Low-resolution full-scan SWS spectra will be obtained for a sample covering all possible kinds of IR features exhibited by S stars. This program should thus provide a library of IR spectral features of S stars which can be used to look how the various features correlate with the luminosity indicator. Mass loss rates, derived by fitting the spectral energy distribution by dust radiative transfer models, can also be correlated to that indicator.