Would one rather store squeezing or entanglement in continuous variable quantum memories?

Given two quantum memories for continuous variables (e.g., the collective pseudo-spin of two atomic ensembles) and the possibility to perform passive optical operations (typically beam-splitters) on the optical modes before or after the storage, two possible scenarios arise resulting in generally different degrees of final entanglement. Namely, one could either store an entangled state and retrieve it directly from the memory, or rather store two separate single-mode squeezed states and then combine them with a beam-splitter to generate the final entangled state. In this paper, we address the question of which of these two options yields the higher entanglement. By adopting a well established descrip- tion of QND feedback memories, and a simple but realistic noise model, we analytically determine the optimal choice for several regions of noise parameters and quantify the advantage it entails, not only in terms of final entanglement but also in terms of the capability of the final state to act as a shared resource for quantum teleportation. We will see that, for 'ideal' or 'nearly ideal' memories, the more efficient of the two options is the one that better protects the quadrature subject to the largest noise in the memory (by increasing it and making it more robust).