So far, all of the people trying difftastic with huge files have been using C or C++ source code. Maybe it's more common in those communities?
(Difftastic will eventually fall back to fast, dumb, line-based diffing if you give it a multi-megabyte source file.)