Tests manipulating the DOM may very well interfere with one another due to improper clean up resulting in lingering DOM artifacts, potentially rendering test results erronous. Running single tests in some kind of sandbox, if possible, would effectively prevent this.