It looks like glDispatchComputeIndirect is still doing a cpu-gpu sync.
While on this topic; is there's a glMultiDispatchComputeIndirect in the ARB pipe?
Preferably such functionality should come with a GLSL built-in gl_DispatchID (like gl_DrawID) and be able to source the number of compute dispatches from a buffer. (like ARB_indirect_parameters)
Retrieving data ...