Double-line eclipsing binaries (DLEBs) have been recently used to constrain the amount of central mixing as a function of stellar mass, with contrasting results. In this work, we reanalyze the DLEB sample by Claret & Torres, using a Bayesian method and new PARSEC tracks that account for both convective core overshooting and rotational mixing. Using overshooting alone we obtain that, for masses larger than about 1.9 M , the distribution of the overshooting parameter, λ ov , has a wide dispersion between 0.3 and 0.8, with essentially no values below λ ov = 0.3 -0.4. While the lower limit supports a mild convective overshooting efficiency, the large dispersion derived is difficult to explain in the framework of current models of that process, which leave little room for large randomness. We suggest that a simple interpretation of our results can be rotational mixing: different initial rotational velocities, in addition to a fixed amount of overshooting, could reproduce the high dispersion derived for intermediatemass stars. After a reanalysis of the data, we find good agreement with models computed with fixed overshooting parameter, λ ov = 0.4, and initial rotational rates, ω, uniformly distributed in a wide range between 0 and 0.8 times the break-up value, at varying initial mass. We also find that our best-fitting models for the components of α Aurigae and TZ Fornacis, agree with their observed rotational velocities, thus providing independent support to our hypothesis. We conclude that a constant efficiency of overshooting in concurrence with a star-to-star variation in the rotational mixing, might be crucial in the interpretation of such data.