On the plus side -- and assuming it's feasible -- it might pose less work for the janitors who get a fair number of considerations requesting "restore original content" (which they can and sometimes do, whilst differentiating the OP from its updated content.

But, TTBOMK, Corion's records, at http://corion.net are the best source for original content and are accessible to anyone who wishes to review a node for possible un-flagged updates. I keep a link in my Free Nodelet for such review. It looks like this:

On the down side, your proposal fails to discourage the abusers who update without warning; especially those who remove their original content in favor of something that's responsive to a reply or to make themselves appear a bit less dunder-headed. A datastamp would do nothing to restore the original... and it's the lack of that which -- IMO -- makes the replies "confusing."

An alternative might be developing a mechanism that "locks down" a node as soon as it receives a reply (thus encouraging a reply to a reply), but even that has its weaknesses, as threads sometimes contain nodes that are tantamount to replies to nodes at the same level.