There is a very revealing paper by Clinger, Hartheimer, and Ost, "Implementations Strategies for First Class Continuations", that provides detailed examinations of various strategies, the code generated to support them, and actual measurements showing the effects of each implementation variation. 

http://www.springerlink.com/content/h5808n962434j275/fulltext.pdf

This paper, along with Marc's recent explanation about the cost of CPS, should go a long way toward explaining why CPS code may be slower than direct form code.

Dr. David McClain
Sr. VP, Embedded Systems
Asyrmatos Inc.
Boston & Tucson
phone:  520-529-2437
cell:  520-390-3995
web:  www.asyrmatos.com
e-mail: dbm@asyrmatos.com