Scene audio volume changes require the scene to be tagged with DEG_id_tag_update(&scene->id, ID_RECALC_AUDIO_VOLUME) (see BKE_scene_update_sound()). Tagging happens in the RNA update function rna_Scene_volume_update(), but that function is not called by the animation system. As a result, animated volume changes are not sent to the audio system.
To resolve this, I've added a depsgraph operation node, which a callback that sets this tag. It depends on the animation evaluation node when the volume is actually animated, as per the AUDIO_VOLUME_ANIMATED flag.