Does the distance between two frozen atoms remain constant throughout the trajectory? If so, the atoms are actually frozen and just translated during the trajectory.
If this is not the case, there are a few ways to freeze atoms in g09 using cartesian coordinates. Although I have no experience in ADMP calculations in g09, you may try to see if any of these options work for your purpose:
1) try changing Geom=ModRedundant with Geom=AddRedundant: the specification of the frozen atoms is the same. From the g09 Manual (http://www.gaussian.com/g_tech/g_ur/k_geom.htm):
"AddRedundant is synonymous with ModRedundant. This option may be used for job types other than optimizations."
2) try changing Geom=ModRedundant with Geom=ReadOptimize or Geom=ReadFreeze: you can find more on this keyword in the Manual (http://www.gaussian.com/g_tech/g_ur/k_geom.htm). In my experience, Geom=ReadFreeze works with ONIOM calculations, even though it's said to be deprecated in the Manual, while Geom=ReadOptimize doesn't work (at least in G09 rev A.02).
3) try using the freeze-code -1 in the molecule specification after the element label (http://www.gaussian.com/g_tech/g_ur/m_molspec.htm).
I had the same problem with QST2: neither Geom=ModRedundant nor Geom=AddRedundant, nor Geom=ReadOptimize, nor even "-1" in molecule specification worked for me. Of course, I kept the same freezing section for both reactant and product.