There's an excellent, comprehensive review article on the entire process of REM sleep:
Vetrivelan,R, Chang, C, Lu,J (2011). Muscle tone regulation during REM sleep:
neural circuitry and clinical significance. Archives Italiennes de Biologie, 149 [DOI] [Free PDF]
Projections from the sublaterodorsal nucleus (SLD) of the Pons in the brainstem are responsible for the suppression of the spinal motor output/movement (for an anatomical schematic, see the article).
Descending SLD glutamatergic projections activating the premotor neurons
in the ventromedial medulla and spinal cord interneurons bring about muscle atonia and suppress phasic muscle twitches in spinal musculature.
Lesions in this area in rats have resulted in REM sleep without atonia.
Many studies have
suggested that the ventromedial medulla (VMM),
which receives glutamatergic inputs from the SLD,
may be the relay site for the pontine inhibitory area
on spinal motor neurons
The SLD axons relay through the VMM, but also directly projects onto interneurons within the spinal cord, directly inhibiting their activation.
Most of the knowledge we have about this subject comes from animal studies, so it will take some time to verify whether these same mechanism and relays are the same in humans.
The area that is lending the most credence to these issues of REM atonia is not sleepwalking as much as REM sleep behavioral disorder (RBD). In this condition, sufferers act out their dreams and generally lose atonia during REM sleep.
Evidence points to the fact that such behavior is often seen decades before such conditions as Parkinson's and Lewy Body dementia.
Ferini-Strambi,L (2011) Does idiopathic REM sleep behavior disorder (iRBD) really exist?
What are the potential markers of neurodegeneration in iRBD? Sleep Medicine 12, S43–S49 [DOI]
They summarized two studies which explored these anatomical changes:
[Diffusion tensor imaging, a type of MRI scan] changes have been reported in the
brainstem (pons, substantia nigra) as well as the fornix, the right
visual stream, and the left superior temporal lobe suggestive of
brain tissue disintegrity
significant decreases of [fractional aniosotropy] in the tegmentum
of the midbrain and rostral pons and increases of [mean diffusivity]
within the pontine reticular formation overlapping with a cluster
of decreased [fractional anisotropy] in the midbrain 
So, the answer is, we don't know for sure, but we have some clues from the animal models as to what stages of the signals to suppress movement are interrupted.
Unger MM, Belke M, Menzler K, et al.(2010). Diffusion tensor imaging in idiopathic
REM sleep behavior disorder reveals microstructural changes in the brainstem,
substantia nigra, olfactory region, and other brain regions. Sleep,
Scherfler C, Frauscher B, Schocke M, et al. (2011) White and gray matter abnormalities
in idiopathic rapid eye movement sleep behavior disorder: a diffusion-tensor
imaging and voxel-based morphometry study. Ann Neurol,69(2),400–7. [DOI]