The study of the distribution of baryonic matter within dark halos enriches our understanding of galaxy formation. We show the radial dependence of stellar baryon-fraction curves derived for 21 lensing galaxies from the CfA-Arizona Space Telescope LEns Survey (CASTLES) by means of stellar population synthesis and pixel-based mass reconstruction. The sample covers a stellar mass range of Ms ~= 2 × 109-3 × 1011 M sun (solar masses) which corresponds to a total enclosed mass range of ML ~= 7 × 109-3 × 1012 M sun on radial scales from 0.25R e to 5R e (effective radii). By examining the Ms and ML dependence on radial distance to the center of each galaxy, we find that there are pairs of lenses on small to intermediate mass scales which approach at large radii the same values for their enclosed total mass but exhibit very different stellar masses and stellar baryon fractions. This peculiar behavior subsides for the most massive lensing galaxies. All the baryon-fraction profiles show that the dark matter halo overtakes the stellar content between 1.5 and 2.5R e. At 3R e most of the stellar component is enclosed. We find evidence for a stellar baryon fraction steadily declining over the full mass range. Furthermore, we shed light on the Fundamental Plane puzzle by showing that the slope of the ML (< R)-to-Ms (< R) relation approaches the mass-to-light relation of recent Fundamental Plane studies at large radii. We also introduce novel concentration indices c = R90/R50 for stellar and total mass profiles (i.e., the ratio of radii enclosing 90% and 50% of the stellar or total mass). We show that the value c = 2.6 originally determined by light profiles which separates early-type galaxies from late-type galaxies also holds for stellar mass. In particular, less massive dark matter halos turn out to be influenced by the distribution of stellar matter on resolved scales below 10 kpc. The ongoing study of resolved baryon-fraction profiles will make it possible to evaluate the validity of star formation models as well as adiabatic contraction prescriptions commonly used in simulations.