I don't tend to go through and do wholesale rewrites. There are two reasons for this; first that I find it more constructive and less problematic to begin a new work with the knowledge of how to improve upon the methods used in the old, and second because I tend to revise ideas in the course of writing the work anyway. I am fortunate that Sibelius keeps an archive of previous file versions and from these I can see how much I have changed a piece in the course of its creation. Sometimes this is quite drastic; whole sections are moved around, deleted, re-written, but this is part of the inital process and not after the piece is fully formed. Mostly, as others have said, the main purpose of any revisions is to solve any technical difficulties or make minor changes to tempi or interpretation markings.
The only major works I can bring to mind that were drastically revised from a workable original are the Violin Concerto and the Fifth Symphony of Sibelius. In the case of the latter, the revisions (adding several important bars to the opening, linking two movements together) arguably elevate the quality of the work and produce a more concentrated argument, but the original versions remain technically accomplished and coherant. This is less so in the case of music such as Vaughan Williams' London Symphony, of which the original version is noticibly inferior to the revised.
Pierre Boulez notes that he considers many of his works revisions of earlier ones, even though the length and content may be very different, and that a composition is never truly finished.