Coded caching (CC) techniques have been shown to be conveniently applicable in multi-input multi-output (MIMO) systems. In a $K$-user network with spatial multiplexing gains of $L$ at the transmitter and $G$ at every receiver, if each user can cache a fraction $γ$ of the file library, a total number of $GKγ+ L$ data streams can be served in parallel. In this paper, we focus on improving the finite-SNR performance of MIMO-CC systems. We first consider a MIMO-CC scheme that relies only on unicasting individual data streams, and then, introduce a decomposition strategy to design a new scheme that delivers the same data streams through multicasting of $G$ parallel codewords. We discuss how optimized beamformers could be designed for each scheme and use numerical simulations to compare their finite-SNR performance. It is shown that while both schemes serve the same number of streams, multicasting provides notable performance improvements. This is because, with multicasting, transmission vectors are built with fewer beamformers, leading to more efficient usage of available power resources.