> Has anyone done any tests to compare the speed with a normal sp?

You can do this by doing something like:

SELECT currentTimeMillis() FROM someTable

and

SELECT CURRENT_TIMESTAMP FROM someTable

you should have approx. 1,000 records there. Then you will see the
performance difference. On my laptop I loose approx. 1 ms for JNI call,
which means that unless we make JNI more efficient (and I'm not sure if that
is possible), this will be suited only for Java procedures with complex
business logic, but not UDFs, where native implementation would be faster.

Roman