We have implemented some significant changes to enhance performance. Significantly, now everything including now services reside in a ReadCommitted cache (not Serializable). Nothing lives in the Serializable Cache. I am in the process of documenting the internal design to make this all clear. At any rate, the TCK runs a lot faster.
Francesco is working on reducing invocations of sbbCreate and postCreate in the implementation. We will also work on controlling thread creations using a ThreadFactory appropriate to the platform you are running on. This could speed things up more. Stay tuned.