It may be worthwil investigating if LTO (together with -Os) could allow further reductions in object size (e.g. allowing inlining across different files).
I don't think that the gcc version of the compiler used by Arduino supports LTO (Link Time Optimization.)
(LTO went into 4.5.x, Arduino runs 4.3.2) It's not clear that the AVR gcc compilers are stable in later releases (4.5 has bugs, 4.6 has bugs; people are looking hopefully at 4.7...)