Now I read this:
In another series of studies (Amaya, Uttal, & DeLoache, under review) we investigated the effectiveness of concrete symbolic objects, known as manipulatives, in helping young elementary-school children learn the procedures associated with two-digit subtraction.... [T]hose children who learned with the manipulatives had trouble transferring knowledge to written versions of the math problems; they did not use what they had learned using the manipulatives to solve written versions of the same or similar problems. Moreover, learning with the manipulatives took almost three times as long as learning with the written method.
American Psychological Association Observer
Granted, that statement applies specifically to subtraction, but if it's applicable to more pervasive uses of manipulatives then I've been wise to caution against their use in most cases.