Quantum communications capacity using direct transmission over length-
L optical fiber scales as $R\ensuremath{\sim}{e}^{\ensuremath{-}\ensuremath{\alpha}L}$, where $\ensuremath{\alpha}$ is the fiber's loss coefficient. The rate achieved using a linear chain of quantum repeaters equipped with quantum memories, probabilistic Bell state measurements (BSMs), and switches used for spatial multiplexing, but no quantum error correction, was shown to surpass the direct-transmission capacity. However, this rate still decays exponentially with the end-to-end distance, viz., $R\ensuremath{\sim}{e}^{\ensuremath{-}s\ensuremath{\alpha}L}$, with
s<1. We show that the introduction of temporal multiplexing---i.e., the ability to perform BSMs among qubits at a repeater node that were successfully entangled with qubits at distinct neighboring nodes at different time steps---leads to a subexponential rate-vs-distance scaling, i.e., $R\ensuremath{\sim}{e}^{\ensuremath{-}t\sqrt{\ensuremath{\alpha}L}}$, which is not attainable with just spatial or spectral multiplexing. We evaluate analytical upper and lower bounds to this rate and obtain the exact rate by numerically optimizing the time-multiplexing block length and the number of repeater nodes. We further demonstrate that incorporating losses in the optical switches used to implement time multiplexing degrades the rate-vs-distance performance, eventually falling back to exponential scaling for very lossy switches. We also examine models for quantum memory decoherence and describe optimal regimes of operation to preserve the desired boost from temporal multiplexing. Quantum memory decoherence is seen to be more detrimental to the repeater's performance over switching losses.