Recent studies on the planet-dominated regime of Type II migration showed that, contrary to the conventional wisdom, massive planets can migrate outwards. Using 'fixed-planet' simulations these studies found a correlation between the sign of the torques acting on the planet and the parameter K' (which describes the depth of the gap carved by the planet in the disc). We perform 'live-planet' simulations exploring a range of K' and disc mass values to test and extend these results. The excitation of planet eccentricity in live-planet simulations breaks the direct dependence of migration rate (rate of change of semi-major axis) on the torques imposed, an effect that 'fixed-planet' simulations cannot treat. By disentangling the contribution to the torque due to the semi-major axis evolution from that due to the eccentricity evolution, we recover the relation between the magnitude and sign of migration and K' and argue that this relation may be better expressed in terms of the related gap depth parameter K. We present a toy model in which the sign of planetary migration changes at a limiting value of K, through which we explore planets' migration in viscously evolving discs. The existence of the torque reversal shapes the planetary system's architecture by accumulating planets either at the stalling radius or in a band around it (defined by the interplay between the planet migration and the disc evolution). In either case, planets pile up in the area 1 - 10 au, disfavouring hot Jupiter formation through Type II migration in the planet-dominated regime.