The difference is the question of the unstressed syllable in the middle of en retard dropping out. This can indeed be done in rapid speech.
For a word like appeler, you can expect even a dictionary to mention the fact that the middle syllable can drop out. In fact, it's normal. But for retard, if viewed in isolation, you couldn't do so, because [ʁt] is not a valid onset.
However, in very fluid speech, sequences like this often break phonological barriers like word edges. That is, the [ʁ] is happy to become the coda of a previous open syllable even if it's in another word, and then the unstressed syllable will indeed drop out. It helps that en retard is such a common collocation that an unconscious parsing as one lexical item is plausible, but even in a phrase like se retrouve par terre the first three syllables could be realized as two: se r'trouve par terre [səʁ tʁuv ...].
For en retard, you can expect to hear the two-syllable pronunciation in quick, familiar speech but all three syllables everywhere else. Here's an example with three and here's an example with two.
As a learner, you can never go wrong with three syllables here. (Whereas for single words like appeler, pronouncing all three might seem like you're spelling it out for someone hard of hearing. That's my impression as a non-native speaker, anyhow, so I invite correction if needed!)