In this proof of concept (currently only tested and working in Chrome 35+) I've split the mp3 in 25 parts by using the unix split command. The moment the first part is loaded then the playback starts immediately and it loads the second. When the second part is loaded then it is appended to the SourceBuffer by using the Media Source Extensions API.
A lot of thanks to Theo for letting me use his awesome track Breathe In.
There is another implementation that I created a year ago which uses only the Web Audio API. You can find it here.