I totally understood that. And again I cannot agree more with you. However, at issue is how Gox perceives this. They're probably thinking from the angle when an auditor or end user standing in front of their face asking for proof that their ABCDE wallet address has been credited for XYZ amount. With a transaction, it has a time, amount and the ABCDE in it, all in one place signed by Gox. With an address, despite the exchange's own claim that it's a one-time and unique, an address by definition CAN be used to send funds to many addresses. Address to transaction is a one-to-many relationship, so you would have to locate the transaction(even there is only one) first in order to get time, amount and the receiving address that the auditor/user need to see, they care the receiving side's proof more than the sending side's.
Ah, I see what you are saying. I suppose from that angle, a re-producable and verifiable hash of a transaction that is actually standardized as part of the protocol might have benefit. At the same time, knowing the tx address is unique can allow you to detect the resultant txid and record that for auditing purposes. A txid is verifiable and permanent after commited to the blockchain, afterall, just not before. So the unique address approach for this is valid and requires no new data to be stored, but wouldn't completely do away with txid's given the type of auditing you suggest MtGox desires.
So, it becomes:
- show user unique sending address as confirmation
- track transaction with sending address
- record txid once transaction is included in the blockchain