We revisit the paradigm of unified dark energy discussing in detail the averaging problem in this type of scenarios, highlighting the need for a full non-linear treatment. We also address the question of if and how models with one or several dark fluids can be observationally distinguished. Simpler and physically clearer derivations of some key results, most notably on the relation between the generalized Chaplygin gas and the standard ($\Lambda$CDM) `concordance' model and on a Jeans-type small-scale instability of some coupled dark energy/dark matter models are presented.