First of all, dual coding is known in virus genomes, where there is a strong size constraint. This is selected for in such circumstances because of that size constraint. As I recall, there is even an example of *triple* coding.
We also know that there examples of double coding in humans. These tend to happen in genes that are not transcribed frequently and thereby have less selective pressure on them. Also, one of the reading frames is transcribed much more frequently than the other.
Now, why this is evidence for ID is unclear. Given the wide variety of ways that transcipription is moderated (including different transcription start points and splicing), it isn't too much of a surprise when more than one frame appears in a stretch of DNA. If both frames encode useful proteins, such would stay around.