Just a simple question: am I wrong or, from a cladistic point of view, any node of the tree should have exactly zero or two descendant, one of which is a terminal node?

For if this is the case the problem becomes quite a trivial one, but your example doesn't match my hypothesis.

Update: forget the second condition, it's bullshit.

But I think that the first one holds, for you can have ((rat,mouse),(human,chimp)) or ((rat,(mouse,(human,chimp)) or (mouse,(rat,(human,chimp)) but it make no sense (rat,mouse,(human,chimp)), particularly if you are checking the variation score independently on each base. For the total score to make sense you should assume one more common ancestor or the score for base X will not be consistent with that of base Y.