So many possible reasons, really. It's possible that there is some setting that is not obvious and is increasing the level of the audio inside the application that is making it exceed 0 dBFS. For instance, if it's a stereo file and the left and right channels are both panned to the center, you could be adding 3 - 6 dB to the summed audio.
Probably more likely is the complicated interplay between MP3 codecs, virtual summing amps, and intersample overages. The MP3 format has to be converted back to PCM to be processed by the software, which means the software is resynthesizing some frequencies. If the encoded level is high enough, the resynthesis could be adding enough to bring it over the top when decoded. The virtual summing amp inside the software could be very short on headroom or there could be another summing quirk like the stereo summing concern above. And even PCM audio that is limited to full scale can be reconstructed into audio that exceeds the full scale level when the full waveform is reconstructed.
When CDs were the primary delivery format (16-bit PCM, essentially), mastering engineers quickly discovered that if they didn't leave at least 0.3 dB of headroom between the peak sample(s) and full scale, some playback equipment would clip, even in the digital domain. I hard limit at -1.0 dBFS because the last thing I want is a digital over and when you start transcoding to MP3 or AAC or whatever, strange things can happen. The original soundtrack to Akira on CD actually has obvious overs right at the beginning of the first track, and there are a few CDs from the early 90s where you can hear them if you listen closely.
