When creating a lag-energy plot the transfer function is normalised so that the maximum sum of any row (i.e., energy) is unity. After careful re-reading of the Cackett et al. (2014) paper, I believe this is what they did.
This is the corresponding lag-energy plot which looks quite plausible. In any case, I think the issue was simply one of normalisation, as you suspected.
When creating a lag-energy plot the transfer function is normalised so that the maximum sum of any row (i.e., energy) is unity. After careful re-reading of the Cackett et al. (2014) paper, I believe this is what they did.