you use zdepth to comp a dof pass. you use your smarts to put one object infront of another in comp.

There are many ways to render something like this the way you want.
you could render the whole lot in separate passes. Then build everything up in comp from the diffuse. this will look like your primary pass, but you will have control over everything. You could use ray type switch node to render some things as mattes but still cast shadows and reflections You could use render layer overrides. you could load up the production mental ray materials and do some reflection and shadow catching.

in short its kind of a long answer, sorry i don't have time to go in to detail.
Maybe someone else will be more helpful.

Out of curiosity, why do you think it wont render together?