The ability of a drug to cross the intestine-blood barrier is a key quantity for drug design and employment and is normally quantified by the permeability coefficient P, often evaluated in the so-called Caco-2 assay. This assay is based on measuring the initial growth rate of the concentration of the drug beyond the cellular barrier but not its steady-state flux through the membrane. This might lead to confusion since, in the case of lipophilic drugs, the initial slope is strongly affected by the retention of the drug in the membrane. This effect is well known but seldom considered in the assay. Here, we exploit all-atoms molecular dynamics and bias exchange metadynamics to calculate the concentration of two lipophilic drugs across a model membrane as a function of time. This allows estimating both the steady-state flux and the initial slope of the concentration growth and comparing Caco-2 and steady-state estimates of P. We show that our computational procedure is able to reproduce the experimental values, although these may differ from the permeability coefficients by orders of magnitude. Our findings are generalized by a simplified one-dimensional model of the permeation process that may act as a roadmap to assess which measure of membrane permeability would be more appropriate and, consequently, whether retention corrections should be included in estimates based on Caco-2 assays.