When I tested those sounds using GMS v135 with NoLifeStory, they were not mp3 data and libmpg123 would choke on them and refuse to play them. When I instead sent the data as is to the sound system as raw samples, it played just fine, although I manually had to specify sample rate and channel count.
Hmm... this makes me think that they were headerless waveform data files (e.g. those .WAV files without their headers) instead. If they were MP3 streams, you should have just gotten some noise instead.
Unless, you specified that the data is encoded as MP3. If you didn't have such an option, I suppose that we can assume that the system played the data as raw waveform data.
If that is the case, then yes... GMS has headerless, uncompressed audio data.