It has to do with subtracting out the 3 dof for the center of mass
motion. The argument
is that if you have pairwise conservative forces (or even
non-pairwise, like 3-body), that
they do not affect the COM motion of the system. Thus there are really 3 DOF
less than the 3N you would otherwise presume. It might be discussed in Allen
and Tildesley, but I can't recall. For reasonably large N, it doesn't
really matter.
Steve