I got the same result under ubuntu linux running dual xeon pIII 500s. It was horribly slow using interpreted scheme. However, when the scheme source was compiled it sped up significantly. It went from about 30 seconds per request to < 1 second per request. I was a bit surprised by that behavior.