I finally got a decent setup to play this: the RetroArch emulator with the 'Beetle GBA' core (Mednafen ported to libretro) with its amazing latency reduction and the following settings:
Hard GPU Sync: ON
Hard GPU Sync Frames: 0
Fame Delay: 0
Audio Latency (ms): 24
Poll Type Behaviour: Late
Run-Ahead to Reduce Latency: ON
Number of Frames to Run Ahead: 2
Runahead Use Second Instance: OFF
