I think something like this would be a very useful feature, but it would be better to have an action which identifies emails as spam, and passes them to a configured handler on the server. Besides reporting to spamcop et. al., it'd be useful to be able to use the moderator actions to train a local bayesian filter.