It is known that plain and computational zero knowledge proof systems are not sequentially composable without auxiliary input (see for example http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.39.7073). The proofs that I've found for this seem to rely on the fact that the transcripts generated by the simulator are only computationally or statistically indistinguishable from those generated by the verifier-prover interaction, so there is some "leeway" for the simulator to work with.
What about perfect zero knowledge proof systems, i.e. proof systems where the prover has unlimited computational resources and for any given verifier-prover pair, there exists a simulator with an identical statistical distribution of transcripts?
Are those sequentially composable without auxiliary input?
