I think Greg Trihus needs to take note of this. This is a very similar issue to what we had with a combined Roman and Arabic script dictionary. When we wanted to print an Arabic version of the dictionary, even though I could sort and format the data the way I wanted it, I could not get the Letter Headers to cooperate - they stayed in Roman script, and jumped all around similar to what you see, A a, I i, E e, A a, E e, etc. The only solution we found was to bump the Arabic script form up to the first Writing System in the database, and then the Letter Headers would follow that Writing System. But that's not going to help you, because the lexeme form is not a separate WS for you, just a separate field.
What I think that means is that the Letter Headers NEED to be extracted from and follow the alphabet of the primary sort order of the alphabet view. There was some reason why Greg was saying this would be difficult (which I can't remember right off hand), but it seems to me that this is a good example of why they need to make it happen. Taking a step back, it makes a lot of sense: If you sort a dictionary by a particular field, the Letter Headers need to follow the Writing System of that field. Otherwise you get all sorts of weird results.