If you mean their Fig. 4 (the one where the two species are indeed mixed), this was nuclear, not mtDNA. And, indeed, the relationships within the acuflavidus/elegans clade are not well resolved at all with these data. (Albeit Cabot's is still clearly closer to Elegant than to Sandwich, hence the tree does conflict with traditional species limits.)The Collinson data is mtDNA too, however, and it has elegans and eurygnathus all mixed in their node - that's what I can't resolve. If Elegant and Cabot's are so distinct from one another, how did Collinson come up with that tree? I get it that the bootstrap values are very very low, but they felt at least good enough to publish it.
They published this tree mainly because it gives good support to a node uniting albididorsalis to bengalensis to the exclusion of maximus. I.e, it supports the non-monophyly of the traditional Royal Tern -- which is what this paper was about.