I have been spending time watching SQL Bits videos lately. So far, the video that I find most interesting is "Designing Highly Scalable OLTP Systems". The material was presented by Thomas Kejser and Ewan Fairweather, who are both members of the SQLCAT team at Microsoft.
Several years ago, I saw a system with the the compilation lock behavior that they describe in their talk. We happened to use the same method that they suggested to remediate the problem with our system. Since the code was in stored procedures, making these changes was not difficult. We measured a transactional throughput increase, but It is always nice to get additional, real-world validation.
I am also intrigued by the "hash partitioning" strategy used to avoid hot pages in tables.