Not that many old things don't work, well in some circumstances even. I'm just having a hard time seeing how throwing away ACID and denormalizing data is "post modern" rather than "back to the future".
Denormalization? How about a materialized view to support common searches to reduce I/O from assembling data.
Aggregates? You could probably abuse and extend entity attribute values to (physically) cluster arbitrary / sparse / repeating field values around a common parent.
It's actually a pretty good article, otherwise, I should probably leave myself a marker to find it. Sooner or later, I'll end up having to maintain one of these things -- long gone contractors will build a "latest and greatest" app using a database system not unlike a 1970 mainframe. There are a number of good work-arounds in the article for dealing with systems with poor indexing capabilities -- a single key field to be filtered.