For those who would like to know more about RBM and why using only one step of CD can work (although more are prefered as the training progress, but you should read it instead of being spoiled), lecture 12 of Hinton's machine learning course on Coursera explains how RBM works and some other stuff. So for those who are intererested, are are the slides + videos which are relevant.
(Actually, video 4 is broken on Coursera, so I'll try to upload it later when/if it becomes available).