Once such a packet is found then the flowgraph has a message sink
and puts this half-baked packet in this queue, together with
possibly another message of a different TYPE (control message) with
some parameters required in the second flow (such as initial phase
estimate, etc).
Have you looked at the m-block design (discussion in list archives)
and work in progress? I realize it's not ready yet, and that you
might want to do something before it is, but it seems like this is an
ad hoc version of the more general scheme.