I've just successfully worked around this bug by specifying the namespace of each cluster. The workaround isn't too onerous, so I wouldn't call this critical.