David - As I understand it, a key component of using AIC and model selection is the declaration of a priori models of interest. Therefore, to put it in terms of writing a paper, all models that show up in your Results section should've already been laid out explicitly within your Methods, and implicitly within your introduction. All models should therefore be reported regardless of their AIC; those <2 are simply the ones that have the most support for being used in drawing conclusions or being integrated via model averaging into further work.
In a similar vein, I believe the AIC gurus frown on using forward- or backward stepwise of all the covariate you happen to have on hand. This creates the opportunity for models with no theoretical backing to slip below the deltaAIC<2 cut off. See: Why do we still use stepwise modelling in ecology and behaviour?. WHITTINGHAM et all 2006. Jrn of Animal Ecology.
