That's true - for instance I believe many of the results of statistical mechanics rely on further assumptions about your choice of macrostates, like the fact th

That's true - for instance I believe many of the results of statistical mechanics rely on further assumptions about your choice of macrostates, like the fact that they are ergodic (i.e. the system visits each microstate within a macrostate with equal probability on average). Obviously exotic choices of macrostates will violate these assumptions, and so I would expect the predictions such a model makes to be incorrect. But ultimately that's an empirical question. Entropy is a more general concept that's definable regardless of whether the model is accurate or not.