Using the rev_sha1 column in the revision table should be more reliable in detecting reverted edits. Currently we determine this by the edit summary and size of the following edit.
Using the SHA might also allow us to detect reverts that happened later on in the revision history.