Lower-fidelity playback systems and environments can actually reveal artifacts in lossily compressed audio. The psychoacoustic encoder makes assumptions about what noises mask other noises - if noises it expects to be present are absent (for instance, if your playback system has a poor frequency response), then low-bitrate noises can be unmasked.

I would not be surprised if additional noise, such as road noise, can also subvert masking phenomena in some way, but I don't know of any studies that tested this.