This paper suggests to leave out input scripts in the tx hashes, so that a tx can not be duplicated with a nonce.
It seems a good idea, can we lobby for discuss this with the core developers?

http://eprint.iacr.org/2013/837.pdf