Scott McNealy's keynote brought back some of the edge. This time instead of Microsoft or IBM, Dell was at the receiving end a couple of times. The Brazilian healthcare project was interesting (2.5M lines of code in 4 months hmm! I will suspend judgement for now).
The best talk of the conference so far was Amazon's web-services talk. Real bits, real experience and real scale. What more could you ask? One interesting tidbit was that about 80% of calls use REST instead of SOAP - most developers are using scripting languages rather than strongly typed languages like C# or Java. Both Java and .NET camps need to ponder this further.
The Spring framework talk was very well delivered and made a lot of sense. However, many (or even most) Java users are going for some kind of J2EE app server (even free ones like JBoss) - so it is not clear how much users will gain from spring on the mid-tier going forward. EJB is co-opting all the "experiements" so given the vendor backing, they seem to have a better shot.
The caching talk started out quite interesting but got less practical as it proceeded. Partitioning in a distributed systems is a very interesting theoretical problem but most implementations simply avoid dealing with it because of costs and complexity. Otherwise, caching opponents call a cache a glorified hash table and the fanatics don't want to admit lack of neral scalability beyond small clusters (except on very specialized workloads.
Finally SDO seems to be dead. As one of my colleagues mentioned, even at their BoF, hardly anyone had used it.