Whatever is representative to how you will be modelling it.

Will you be populating the model based upon Petrophysically derived porosity and a poro perm relationship? Then you need to ensure that the RCA you use represents the same porosity average to that which your tool resolution would yield at that location. If your sat/height function relationships is in turn a function of permeability (ie simple J function), then what permeability are you using? Klinkenberg corrected air perm on cleaned core? Perm of core at residual saturation? It all matters. For perms of sub 1mD, actual permeability in presence of residual water can be 10-100x less than to that of air.

You need to sit down and think it all through to ensure you are consistent, and that in the final analysis your Porosity x (1-Sw) yields appropriate hydrocarbon volumes, and that K x Kr yields appropriate productivity for your various phases under reservoir conditions.