It seems to me that a scripting engine only needs to register with the BSF engine once. I think that JudoScript registers with the engine every time a script is invoked. In order to avoid this situation, I added a static variable to track the bsfStatus. In the test case that I was executed, I was able to reduce calls to Class.forName() from 8,061 to 43 and the created objects from 649,620 to 162. This also saved 2.6% of execution time.

    public static String bsfStatus = "UNKNOWN";

    public static void registerToBSF() {
        if (bsfStatus.equals("UNKNOWN")) {
            try {
                Class[] params = new Class[] {String.class, String.class, Class.forName("[Ljava.lang.String;")};
                Method m = Class.forName("com.ibm.bsf.BSFManager").getMethod("registerScriptingEngine", params);
                Object[] vals = new Object[] {"judoscript", "com.judoscript.BSFJudoEngine", new String[] {"judo", "jud"}};
                m.invoke(null, vals);
                bsfStatus = "AVAILABLE";
            } catch (Exception e) {
                bsfStatus = "UNAVAILABLE";
            } // if BSF is not there, so be it.
        }
    }

Using JProbe, I noticed that my application was repeatedly compiling the same JudoScript scripts. I added a cache for compiled JudoScript scripts which reduced the calls to ParserHelper.parse():

Not a bad first optimization for our first use of the JProbe profiler.