We investigate the effect of different doping schemes on the broadband magnetic losses and their temperature dependence in Mn-Zn ferrites. CaO, Nb2O5, ZrO2, and SiO2 are added with increasing proportions to TiO2-doped prefired powders and, after sintering at either 1275 °C or 1300 °C, the obtained ring samples are tested versus frequency f (DC-1 GHz) and peak polarization Jp (2 mT – 200 mT) up to T = 160 °C. Appropriately enhanced impurity contents are shown to induce further decrease of the energy loss in materials already prepared for best performance at high temperatures (140 – 160 °C). This behavior can be hardly ascribed to the impurity-related increase of the electrical resistivity brough about by extra-doping, being it rather connected to a corresponding monotonical decrease of the effective magnetic anisotropy < Keff > with T. The decreasing anisotropy makes the balance between the contributions of domain wall (dw) displacements and reversible rotations to the magnetization process evolving in favor of the latter. The energy loss correspondingly develops with frequency and peak polarization in a complex fashion, according to the specific dissipative mechanisms sustained by the spins precessing either inside the moving walls or in the bulk. A dividing line in the (Jp − f) plane is identified, which separates dominant dw- and rotation-generated losses. It moves downward (i.e. lower f) with increasing temperature, the higher T the lower the frequency at which the rotations, theoretically assessed via the Landau-Lifshitz equation, supersede the domain wall contribution. Once accomplished, however, the transition to rotations can lead, according to the theoretical model, to higher losses when moving to higher temperatures. Following the experimental trend of the complex resistivity versus frequency at different T values, the calculations and the experiments show that eddy currents start to contribute to the energy loss, in the 5 mm thick ring samples, around a few MHz, accounting for about 50 % of measured loss beyond some 50 MHz. The chief dissipative process at applicative frequencies and induction values is therefore identified with spin damping, to which the generalized loss decomposition method can be applied.