Using hardware acceleration & Solaris crypto with Java has shown to be less than optimal due to java memory allocation issues and the number of layers to go through the JNI to get to the OS library to do a simple operation. When prototyping a T3 JCE provider, which optimized the solaris side, showed no performance gain because of memory management.
While those java memory management issues may have improved, the reality is that we need to have a quicker and more direct way to get to the OS libraries. It is the hope that by having a more direct access to the library, that we can eliminate some of the memory and other java calls that are needed use JNI and leave that work to the OS library were it's faster. The quicker access also allows us to get better performance from smaller byte sizes of crypto/message digest operations.
This is of particular importance for the new crypto hardware acceleration in T4 and the availability of the libsoftcrypto/libmd libaries that allow quick, lightweight access. This combination could be a great perfomance improvement for Solaris/SPARC, potentically record numbers (?).