I'd go with option C, since that's fundamentally the purpose of using intermediate code (another one is to perform high level optimizations).
For instance, GCC has got two intermediate langauges - "GIMPLE" and "RTL". The front-end translates source into GIMPLE on which high-level optimizations are done, and then lowers to RTL on, which low-level target speciifc optimizations are performed. One can leverage the optimizations in GCC by writing a front-end and a back-end for GCC as opposed to re-writing the "middle-end". In a nutshell by using an IR, writing m front-ends and n back-ends is sufficient as opposed to writing m * n compilers.
