I agree that the key issue is the hypothesis being tested.  If you wish to test whether an intervention on one hemisphere causes a measureable change in that hemisphere, then comparing an appropriate untreated or sham-treated animal certainly gets around the problem that your intervention may well (in fact, likely) affect both sides.  If, however, you are studying a naturally-occuring event which is rare (for example, Rasmussen's encephalitis or hemimegencephaly in humans), a comparison of one side versus the other might be the most appropriate (if you were looking at, say, metabloism by PET or white matter volume by MRI).  In such a case, the comparison of one side against the other allows within-subject statistical power, and automatically controls for age, gender, and many genetic factors.

For the situation your question suggests, comparison with another individual brain seems most appropriate.

Definitely I reckon the use of different animals for the control experiment. Many of parameters have been shared through the ventricular access between the hemisphere so if one use the contralateral hemisphere as the control one then it is more stressful than that in normal mice. Therefore many study parameters can not be achieved correctly using the same animal as treated and control group. Stress during the experiment, drug treatment and animal handling can not be neglected while doing animal experiment. So I better suggest to use separate animals for the different experimental group. However certain parameter are same among sham and the contralateral hemisphere of experimental animals. Such as microglia expression and activation state after focal cerebral ischemia. Generally old papers use same animal as treated and control hemisphere.

Completely agree with Bhakta - you can never be sure that an intervention's effects will stay localised to one hemisphere. In the pain field, many models use a unilateral peripheral nerve injury, and effects can sometimes be seen in both the contralateral limb, and in both hemispheres.

In addition, as you mention, there are functional differences between the hemispheres, so depending on what your outcome measure is, there may well be lateralisation effects. 

The safest thing to do is include at least 3 groups - test, sham, and naive - that way you can be clear on what effects are due to the model/intervention, and which are natureal hemispheric effects.

I agree that the key issue is the hypothesis being tested.  If you wish to test whether an intervention on one hemisphere causes a measureable change in that hemisphere, then comparing an appropriate untreated or sham-treated animal certainly gets around the problem that your intervention may well (in fact, likely) affect both sides.  If, however, you are studying a naturally-occuring event which is rare (for example, Rasmussen's encephalitis or hemimegencephaly in humans), a comparison of one side versus the other might be the most appropriate (if you were looking at, say, metabloism by PET or white matter volume by MRI).  In such a case, the comparison of one side against the other allows within-subject statistical power, and automatically controls for age, gender, and many genetic factors.

For the situation your question suggests, comparison with another individual brain seems most appropriate.

Here is an example of lesions on one side of the brain in rats influencing the contralateral side:

http://www.ncbi.nlm.nih.gov/pubmed/1626311

Unless directly proved otherwise (through personal studies or reliable references), you'd better expect contralateral effects following unilateral manipulations. Examples are many, as suggested by previous respondents above. You may find others in some of my papers. In conclusion, my advice is don't ever use lightly "the other side" as a control.

Our previous histochemical and immunohistochemical study was aimed to reveal the trauma-induced neuronal injury in the brain of rats with spinal cord (SC) hemisection. At 3 weeks after SC hemisection, the neurodegeneration was observed on the one hemisphere (to determine exactly ipsilateral or contralateral side was impossible because the brain sections used by us were free- floating sections); besides, an asymmetrical localization of the hypothalamic neuroprotective proline-rich polypeptide - 1 (PRP-1) was revealed in the paraventricular and supraoptic nuclei: no PRP-1-immunoreactivity was exhibited at the affected side of both nuclei. Daily intramuscular administration of PRP-1 for 3 weeks significantly increased the number of PRP-1-immunoreactive cells and varicose nerve fibers in the investigated cell groups.

Silva - just a comment on your mention of free-floating sections making hemispheric identification difficult, I tend to make a shallow cut down along one hemisphere to allow differentiation