I haven't done it, but I hear its excellent.
That said, looking back, I'm not sure it encourages the best coding style or the best idioms. It's perfect for getting your head around the paradigm though!
I'm currently writing an analogous tutorial for implementing Prolog in Haskell, but it's on a bit of a hiatus thanks to some school and research related deadlines coming up. Hopefully people are interested in something like that :).
Interestingly enough, the Prolog interpreter is actually shorter than the simple Scheme one--largely because we can take advantage of the list monad for the backtracking. I think it's still a very cool exercise.
I wrote it as a project work in an Artificial Intelligence class. It very closely resembles the logic programming language implemented in "Structure and Interpretation of Computer Programs".
I'm sorry there isn't any documentation for it.
Here's a little example of what it looks like, from examples/royalfamily.slo:
(siblings ?x ?y) <-
(parent ?x ?parent) & (parent ?y ?parent)
(grandparent ?x ?y) <-
(parent ?x ?z) & (parent ?z ?y)
(cousins ?x ?y) <-
(grandparent ?x ?gp) & (grandparent ?y ?gp) & !(siblings ?x ?y)
(I'm only being partly tongue-in-cheek - that's mentality behind idiomatic Lisp).
"Scheme" has many different dialects as well. Racket is _a_ Scheme, Edwin interprets MIT's version of Scheme, then there is Gambit Scheme. There are certainly other Schemes, and the implementation of Scheme on Haskell isn't going to be the same as the other dialects listed here.
There are quite a few examples of Prolog-to-Lisp interpreters, but no one would argue that any of these _are_ Prolog.
Once you are comfortable with Haskell the code makes much more sense and is a great starting point for writing an interpreter of your own.