This study investigates the blade loads on a model tidal turbine subject to unsteady gust forcing in the form of uniform small-amplitude oscillations in the axial inflow velocity. The validity of industry-standard 2D strip-theory models for calculating unsteady hydrodynamic loading on 3D rotor geometries is evaluated, by comparing predictions by the Theodorsen function to numerical Reynolds-Averaged Navier Stokes (RANS) simulations and three-dimensional inviscid vortex lattice modelling (VLM). The results show that the 2D function captures neither the trends nor the magnitudes of the unsteady turbine loads. The inviscid VLM corresponds well to the unsteady RANS simulations, suggesting that 3D wake effects are highly consequential for the unsteady loads. The primary non-dimensional parameters determining the unsteady load magnitudes are identified, and it is observed that industrial turbines are likely to experience operation at the "critical forcing period" corresponding to peak unsteady load amplitudes in realistic wave conditions. These peak loads exceed those predicted by assuming quasi-steady conditions, which is otherwise often considered to give conservative load estimates. We demonstrate the potential hazard in applying quasi-steady tip-loss corrections and induction factors in unsteady flow conditions, as this can substantially under-predict the loads, and show that the shortcomings of strip-theory models stem from their lack of 3D unsteady wake effects. These outcomes have implications for the evaluation of peak and lifetime loads on tidal devices, and for any rotor application which relies on 2D strip-theory for load predictions.