If the compound could be ground to molecular size and if the resulting sample were kept in a 'pile' so that these molecules were in contact with each other, then the molecules would tend to bond together, at least as they would in a liquid and there would be no grains to speak of. If however, the temperature of the 'pile' was maintained below the compound's melting point, the molecules that are in contact with each other would again bond, but the bonds would favor those that would reform the crystal structure it had before grinding. This process would continue (implying that there is some kind of time constant involved with this grain growth) to add molecules to the grain up to a point where further bonding would not be energetically favorable due to competing sources of energy, primarily thermal I would think.
The fact that there is a grain growth rate implies that the method of sample preparation must be considered as well. If this method were continuous (like two blocks of the sample in continuous motion sliding in contact so that the interstitial friction provides the crushing energy) there would be some kind of 'dynamic' grain size established. Compare this to an intermittent grinding action as would be seen when using a traditional ceramic mortar/pestle or piston/hammer type mortar in a laboratory. Here the system will experience periods of no grinding which would allow more time for stable grains to reform, resulting in a larger final grain size. For this question, we are primarily interested in the latter scenario, however the time dependence is important to my application so that any related discussion would be welcome.
This re-bonding would also be affected by the environment the ground compound was in; if it were reactive (e.g. like fluorine or chlorine) chemical processes would be involved, which are to be avoided here, in short, we do not consider reactive environments. If it's in water, then issues of hydration and/or solubility become important for many compounds, again to be avoided in this discussion (but are also important in my final application). Two acceptable environments to consider here would be either a vacuum (provided the compound would not `outgas'), or an inert gas such as helium or neon. I would expect the latter to contribute thermally to the grain size limit more than a vacuum would due to the gas' contribution to the thermodynamics of the system (hopefully, this effect can be ignored).
With these considerations, the question can be restated as:
What material and thermodynamic properties of the compound are predominant in controlling the resulting minimum grain size and how would these properties be employed to estimate the number of molecules in such a grain of sample powder?