We study the physics potential of future long-baseline neutrino oscillation experiments at large $\theta_{13}$, focusing especially on systematic uncertainties. We discuss superbeams, \bbeams, and neutrino factories, and for the first time compare these experiments on an equal footing with respect to systematic errors. We explicitly simulate near detectors for all experiments, we use the same implementation of systematic uncertainties for all experiments, and we fully correlate the uncertainties among detectors, oscillation channels, and beam polarizations as appropriate. As our primary performance indicator, we use the achievable precision in the measurement of the CP violating phase $\deltacp$. We find that a neutrino factory is the only instrument that can measure $\deltacp$ with a precision similar to that of its quark sector counterpart. All neutrino beams operating at peak energies $\gtrsim 2$ GeV are quite robust with respect to systematic uncertainties, whereas especially \bbeams and \thk suffer from large cross section uncertainties in the quasi-elastic regime, combined with their inability to measure the appearance signal cross sections at the near detector. A noteworthy exception is the combination of a $\gamma=100$ \bbeam with an \spl-based superbeam, in which all relevant cross sections can be measured in a self-consistent way. This provides a performance, second only to the neutrino factory. For other superbeam experiments such as \lbno and the setups studied in the context of the \lbne reconfiguration effort, statistics turns out to be the bottleneck. In almost all cases, the near detector is not critical to control systematics since the combined fit of appearance and disappearance data already constrains the impact of systematics to be small provided that the three active flavor oscillation framework is valid.