The replica exchange/parallel tempering method and its variations offer the hope of improved sampling for many challenging problems in molecular simulation. Like all sampling methods, however, the effectiveness of replica exchange is highly dependent on the specific physical system being studied. We have carried out large scale temperature replica exchange (T-REMD) simulations of peptides and proteins in explicit solvent and encountered a number of issues that limit the effectiveness of the method in sampling biomolecular conformations. In particular, our results suggest the need to either replace or augment the temperature variable with an alternative extended variable, such as Hamiltonian scaling.