I second the recommendation to experiment.
My white paper on signals and noise is on the community page of the Bottlehead site, and gives some ideas about setting levels for optimal balance between overload and noise floor. But it does not much address the small-signal nonlinearities that characterize some solid state (op-amp) circuits and all digital approaches. I'm not familiar with the details of the Squeezebox analog circuitry, i.e what kind of amplifiers are used, or how the level control is implemented, so I don't know what would be best for it - hence I also recommend trying various combinations to see what works best. Please post back if you do experiment, it will help many others as well!