I haven't done any extensive tests on timing, but there are two sources for delays:
1) Time to open file - this is in the order of 10s of milliseconds (typically 10 to 50 milliseconds).
2) MP3 files are notorious (depending on the software used to compress them) for having massive amounts of data in the header (for older tags, and other header information) that is pushed through the decoder.
We can't do anything about #1. It takes time to open the file. (Although, I'm looking into a new way to queue files with the "PC N
If you want to eliminate the time wasted with the MP3 headers, you can use PCM WAV files instead, but because of the bandwidth required for PCM files, you'll be limited to mono 22kHz or stereo 11kHz audio. If you're just doing speech, mono 11kHz or 22kHz is fine (e.g. AT&T Labs Natural Voices
only produces mono 16kHz PCM data).