We critically survey the paradigms of molecularly targeted cancer therapy in light of systems-level observations on tumor robustness and resilience. Multipronged attacks covering various clinical fronts such as angiogenesis, apoptosis and cancer progression, among others appear to be favored approaches at present, yet the enhancement of side effects has become a serious concern. In this regard, we argue that a departure from the yet untested single-target paradigm in favor of the notion of 'selective nonselectivity' may be necessary, but will ultimately require a rational control of specificity to curb side effects. This control may be achieved by drug redesign guided by known selectivity-promoting molecular features.