The numerical calculations are usually always carried out in the grand canonical ensemble, so you assume your system of $N$ particles is connected to a large particle reservoir. The chemical potential is fixed, which ensures $\langle N \rangle$ is fixed.
This is of course not physical, but in the thermodynamic limit everything tends to the same anyway.
But what really happens is that the total number of particles (system + reservoir) is conserved because of baryon number conservation. So the fluctuations in atom number for the superfluid phase would be balanced by equal and opposite fluctuations in atom number in the reservoir. They have actually proven this experimentally for partially and fully condensed Bose-Einstein condensates.