If your m-files are scripts running independently of each other, my choice would be to use a .mat file. Save all the variables in a .mat file and then using load with an output argument, load only the variables needed by a particular script into the workspace.
If your m-files are all functions running in your workspace at the same time, you can pass the variables as arguments as needed by the functions and then return new values for them as function outputs (or use the much less-desirable option of declaring them as global variables).
It all depends on what your m-files are doing, and whether they need the variables serially or simultaneously.
