Superluminous supernovae (SLSNe) are massive star explosions that are too luminous to be powered by traditional energy sources, such as the radioactive decay of 56 Ni. These transients may instead be powered by a central engine, such as a millisecond pulsar or magnetar, whose relativistic wind inflates a nebula of high energy particles and radiation behind the expanding supernova ejecta. We present three-dimensional Monte Carlo radiative transfer calculations of SLSNe which follow the production and thermalization of high energy radiation from the nebula into optical radiation and, conversely, determine the gamma-ray emission that escapes the ejecta without thermalizing. Specifically, we track the evolution of photons and matter in a coupled two-zone ("wind/nebula" and "ejecta") model, accounting for the range of radiative processes including (typically multiple generations of) pair creation. We identify a novel mechanism by which γγ pair creation in the upstream pulsar wind regulates the mean energy of particles injected into the nebula over the first several years after the explosion, rendering our results on this timescale insensitive to the (uncertain) intrinsic wind pair multiplicity. To explain the observed late-time steepening of SLSNe optical light curves as being the result of gammaray leakage, we find that the nebular magnetization must be very low ε B ∼ < 10 −6 − 10 −4 . For higher ε B synchrotron emission quickly comes to dominate the thermalized nebula radiation, and being readily photoelectrically absorbed because of its lower photon energies, results in the supernova optical light curve tracking the spin-down power even to late times ∼ > 1 year, inconsistent with observations. For magnetars to remain as viable contenders for powering SLSNe, we thus conclude that either magnetic dissipation in the wind/nebula is extremely efficient, or that the spin-down luminosity decays significantly faster than the canonical magnetic dipole rate ∝ t −2 in a way that coincidentally mimicks gamma-ray escape. Our models predict a late-time flattening in the optical light curves of SLSNe, for which there may be evidence in SN 2015bn.