Apparenlt "High Performance Data Access" at JavaOne means "go pay Oracle for TopLink." Seeing as the guy doing the presentation is the product manager for TopLink, I am not surprised. I am annoyed at how TopLink specific all his examples are though.
Interestingly, they use object caching not field caching just like we do (which we got pinged on recently). We do need to get our two-level cache working though.
I also need to get the aspect based dirty tracking and isolation working. The compile requirement will be annoying, but the capabilities of the tool would be awesome performance wise.