We discuss the limitations of the applicability of the Lifshitz formula to describe the temperature dependence of the Casimir force between two bulk lossy metals. These limitations follow from the finite sizes of the interacting bodies. Namely, Lifshitz's theory is not applicable when the characteristic wavelengths of the fluctuating fields, responsible for the temperature-dependent terms in the Casimir force, is longer than the sizes of the samples. As a result of this, the widely discussed linearly decreasing temperature dependence of the Casimir force can be observed only for dirty and/or large metal samples at high enough temperatures. This solves the problem of the inconsistency between the Nernst theorem and the "linearly decreasing temperature dependence" of the Casimir free energy, because this linear dependence is not valid when $T \to 0$ ; Comment: 8 pages