Bringing up old topic just because this became personally relevant, and perhaps useful for someone else as well.
You can do a NumForce calculation simulating the frozen state of the atoms by setting their mass to "infinite", and running a normal NumForce (without -frznuclei). Edit the control file and add very heavy masses to the frozen atoms. For example, assuming carbons 1 and 2 were frozen during optimisation:
c 1-2 \
basis =c def2-SVP \
jbas =c def2-SVP \
mass =99999999999.9
c 3-6 \
basis =c def2-SVP \
jbas =c def2-SVP
You'll get the same results as with NumForce and -frznuclei, except that the inactive modes will have (almost) zero frequencies. NumForce with -frznuclei sometimes produces some unphysical very high frequencies, these will also be "zero". The total number of vibrations, including translation and rotation, will be 3N.
This still doesn't make freeh happy, though. Most other programs ignore very low frequencies, but freeh includes them for computing thermodynamics, and will blow up. I have used Viewmol for computing thermodynamics from similar calcs (it ignores freqs below 10 cm-1, IIRC), but I'm sure there are other alternatives.
Note that, as Arnim mentioned, this might not be physically completely sound. But if you are interested in relative enthalpy/entropy contributions for systems with, say, the same frozen atoms, I don't see why this wouldn't be a good approach. If someone knows different, speak up :-)