Extremely interesting talk today about making a highly scalable VM for java that can handle a high number of threads far better than the current VMs out there.  What's better than that?  It runs your code better without requiring you to change any of your code at all.  How does it work?  At a high level it allows the VM to actually run your synchronized code in parallel across all running threads, and you only pay a penalty if there would have been data contention between two threads.  In the common case there will be no problem and your thread will execute the critical section *without* acquiring a lock.  And if there is a conflict it can be detected with existing techniques, and standard transactional rollback methods can come to the rescue here.  Apperantly this idea comes into play in databases and is used to allow scalable performance on machines with a high number of CPUs (like 128+).  It sounds very interesting, but i'm extremely confused how it deals with those pesky issues maintaining atomicity or consistancy.

With their testing they've found that the can an amazing level of scaling where in previous systems you could hit a wall extremely quickly.

It's enormously interesting, and far too complicated to go into depth fully in this post.  So if you want to read more google a bit.  From that, i found this very interesting paper on the the topic.