The performance of nano-scale energy conversion devices is studied in the framework of state-space models where a device is described by a graph comprising states and transitions between them represented by nodes and links, respectively. Particular segments of this network represent input (driving) and output processes whose properly chosen flux ratio provides the energy conversion efficiency. Simple cyclical graphs yield Carnot efficiency for the maximum conversion yield. We give general proof that opening a link that separate between the two driving segments always leads to reduced efficiency. We illustrate these general result with simple models of a thermoelectric nanodevice and an organic photovoltaic cell. In the latter an intersecting link of the above type corresponds to non-radiative carriers recombination and the reduced maximum efficiency is manifested as a smaller open-circuit voltage.