I think it may also be the case that (old-school, non-concatenated but generated) synthesized speech, with practice, is more intelligible at higher speeds than natural speech. The sounds are more distinct and consistent.
The speech synthesis for screen readers is heavily tailored for that job (I remember the emphasis put on the new voiceover voices on Mac many years ago, and they apparently were noticeably superior to people who used them as screen readers at the high speed settings)