Cache-aided MIMO communications aims to jointly exploit both coded caching~(CC) and spatial multiplexing gains to enhance communication efficiency. In this paper, we analyze both the achievable degrees of freedom~(DoF) under linear processing constraint and the finite-SNR performance of a MIMO-CC system with CC gain \(t\), where a server with \(L\) transmit antennas communicates with \(K\) users, each equipped with \(G\) receive antennas. We first demonstrate that the enhanced DoF of \(\max_{β, Ω} Ω\times β\) is achievable with linear processing, where the number of users \(Ω\) served in each transmission is fine-tuned to maximize DoF, and \(β\le \min\big(G, \nicefrac{L \binom{Ω-1}{t}}{\big(1 + (Ω- t - 1)\binom{Ω-1}{t}}\big)\big)\) represents the number of parallel streams decoded by each user. Then, we propose a new class of MIMO-CC schemes using a novel scheduling mechanism leveraging maximal multicasting opportunities to maximize delivery rates at given SNR levels while still adhering to linear processing constraints. This new class of schemes is paired with an efficient linear multicast beamformer design, resulting in a more practical, high-performance solution for integrating CC in future MIMO systems.