I've recently open-sourced a little python library I've been using a lot in my work. It provides an easy interface for storing many numpy arrays in shared memory. I use it when running distributed neural network training algorithms, as it allows for global parameter storage with asynchronous updating from independent processes. It could be used for a number of other distributed algorithms applications, though.
Check it out here: https://github.com/mharradon/SHMArrays