A "rainbow table" works in a somewhat similar manner, but is based on doing a full calculation then saving only certain starting points. I think what they are doing is actually more like creating a regular expression that matches all words a particular Levenshtein distance from the target.

Here's more about what how they are doing it: http://blog.notdot.net/2010/07/Damn-Cool-Algorithms-Levensht...

This article has a very good overview of the options: http://stevehanov.ca/blog/index.php?id=114