Don Park is more down on Atom than I am:
For example, what is the point of requiring all entries to have a unique identifier when feed clients have no incentives to reject feeds that violate that requirement? Filtering of duplicate entries from multiple sources is a quality of service problem that is best dealt with in a separate opt-in specification. Clients and servers that support the opt-in spec will be more popular if resulting differences in user experience and distribution are significant enough. If not, there was no need for it anyway.
The reality is, aggregator implementors are not going to charge out and ponder the spec in great detail. We are going to do a sanity check on Atom 1.0 feeds as they appear, tweak existing code as necessary, and move along. In BottomFeeder, RSS feeds, RDF feeds, and Atom feeds all end up being pushed into the same object model - from the end user's standpoint, a feed is a feed, and whether it's Atom, RSS, or RDF just doesn't matter.
The net value of the entire Atom exercise? Very close to zero.