Glycine nucleation simulations

I am trying to run nucleation simulations of glycine in water. To generate a supersaturated solution, a solution that will crystallize, I need to know the solubility. Luckily for me, this data exists. Now at 1 atm, 324 K, glycine's solubility is 0.338 g/ml water.

My question is how to determine the appropriate box size for a given simulation, if I specific the number of glycine molecules. I posted a question similar to this before,in the CCL mailing list, and was told to just assume ml water is ml of simulation box. For instance, if I want 1024 glycine molecules, whose molecular weight is 75.066 g/mol,

  (1024/6.0223*10^(23))*(75.066) = weight = 1.27*10^(-19) g
weight/0.338 = ml simulation box = 3.78*10^(-19) ml
box = ml simulation box^(1/3) = 72.28 angstroms


I guess my first question is if this procedure appears to be correct. Specifically, all experimental data for solubility is clearly gathered at 1 atm pressure. However, if I fix the pressure, my understanding is that there is only one equilibrium concentration that the MD will take me to. If I think of the number of DOF's here, temperature is fixed, pressure is fixed, and the mol fraction is fixed, so I don't get to choose the box and hence the solubility. If I work in the NVT ensemble, which I am leaning towards, I don't know if I will actually be supersaturated because the simulation is no longer ran at 1 atm. Is there another definition of solubility that I should be using for MD?

The above scenario is more confusing to me when I think of simulating a "bulk" crystal. If I were to simulate a glycine crystal consisting of 1024 molecules, the above method would tell me a simulation box of 72.28 angstroms and a temperature of 294 K would mean the crystal wouldn't dissolve (assuming nanoscale stability is similar to bulk stability). However, if I am simulating a bulk crystal, why would the box size matter to the crystal? It should always see a "sea" of solvent, and I thought I should only be concerned with making sure the box is large enough to prevent the cluster from seeing its own periodic images.