To rationalize coherence and mechanochemical aspects of proteins acting as molecular machines, a plasmon concept for dealing with protein nonequilibrium dynamics is introduced and tested with respect to thermodynamic consistency. A stochastic optimum-control theory for protein conformational diffusion is developed and the corresponding stochastic Newton's second law derived for optimum-controlled conformational diffusion in proteins. The plasmon concept is shown to be consistent with this theory, in that optical plasmons can pump entropy out of (or into) the protein, decreasing (or increasing) its conformational diffusion and, at the same time, help decrease intra- and intermolecular friction, as well as (potentially) break the symmetry of the latter. Instead, acoustic plasmons may break the spatial symmetry of a protein's "potential of mean force", thus converting it into an effective Brownian ratchet potential by applying quasistatic deformational corrections to the former. These concepts seem to be of rather general applicability and might also be useful when studying, for example, intercalation of cationic dyes into DNA duplexes, positively charged oligopeptide transduction through cell membranes, or even DNA translocation through nanopores.