For the simple case where the Java applications are run in the voice response node, the Java version of the application uses more of the processing capability than the highly-optimized state table version. However, the increasing performance of pSeries computer means that the specific processor should not be a limiting factor. The memory used in the system increases only by the amount required for the Java process.
If you run the applications in a distributed model (that is, in a separate node from the voice response node), you need more processing power. However, in most cases, the added flexibility and redundancy you achieve from the distributed model is more important than the slight performance advantage you can get from running your applications in the voice response node.
Running CCXML applications requires a relative increase of 25 to 30% in processing power compared with that required to run VoiceXML 2.1 applications, depending on the complexity of the applications. For example, if 50% of the available processing power is used by a VoiceXML 2.1 application, 62.5% is used when a CCXML application is also run on the same machine.