The high computational cost of abduction has limited the application of this powerful and expressive formalism to practical cases. SCIFF is an abductive proof procedure used for verifying the compliance of agent behaviour to interaction protocols in multi-agent systems; SCIFF has been integrated in SOCS-SI, a system able to observe the agent interaction, pass it to SCIFF for the reasoning process and to display in a GUI the results of the SCIFF computation. In order to assess the applicability of sciff and SOCS-SI to practical cases, we have evaluated qualitatively and experimentally (not yet formally) their computational behaviour, concerning limitations and scalability. In this paper we show the results of the analysis.