The ylim parameter in plot_evoked_topo is None by default and for this default, the channels have different ylim. I think this is a bad default because:

1. It does not allow comparison across channels (e.g., lateralization effects)
2. It does not allow you to easily spot bad channels

Instead, the ylim should be determined based on all the channels. One option is to copy the defaults from xplotter which are based on expected physical values of the quantities for evoked responses. The other option is to use the MAD or some such thing so that you get reasonable values even when bad channels are present.