...Your idea would be: send packet at time T, wait for the first copy to arrive, count the number of additional copies, in other words, arrive in the interval [T+OWD, T+OWD+delta]. That could work too but makes the definition somewhat incompatible with the loss definition.
I don't understand the value of a relative waiting time as described above. It might be worth thinking about characterizing the time between arrival of the first packet and subsequent copies, but it would clearly be a secondary sort of metric w.r.t. the duplication itself.
Al