Why only the leader sings
A marble race generates a lot of impacts. If every one of them advances the melody, the song is played by the pack as a whole, which means it is played by nobody in particular. The result is a texture rather than a tune, and it sounds the same whoever is winning.
Handing the melody to the current leader changes what the audio is for. It stops being ambience and becomes a readout: the song is a property of first place, and a lead change hands it over mid-phrase. Viewers notice the handover before they have consciously read the leaderboard.
It also solves a practical problem. A single voice advancing one note per impact keeps the melody recognisable at any pack size, where a dozen balls all advancing it would scramble the tune within seconds.

