Okay, I have to admit that I was looking at this problem incorrectly, and thought that I had a solution.
For some reason I thought that you only had 120/240V _single phase_ loads to supply, and I was thinking about ways to use your 480:240 delta-delta transformer as a single phase 480:120/240 transformer. (The basic idea: leave one supply phase open and then treat the secondary terminal associated with the open primary terminal as the neutral.) Using this approach, I think that your 300KVA delta-delta transformer could be effectively used as a 200KVA single phase transformer.
Now I realize that you need to supply 240V three phase loads as well.
I think that your original proposal (using some sort of transformer to derive a midpoint tap on one of the delta legs) has merit. This would be the single phase equivalent of using a zig-zag transformer to derive a neutral point on a delta secondary. It seems to me that if you connected a suitable center tapped autotransformer from phase A to phase C you would derive a useable neutral.
However given the size of the loads, I believe that you would be better off with two separate transformers in a more standard configuration rather than trying to add an autotransformer to a delta. I believe that the unbalanced loading that you describe would push the A and C legs very near to full capacity, possibly beyond. Then you have the problem of sizing the autotransformer.
As a first pass, I believe that you would need the full 115 KVA capacity for your neutral deriving transformer. If I had to guess, a half sized weird transformer (as an autotransformer, you only have losses in one coil, so an X KVA isolation transformer can serve as a 2X KVA autotransformer) would not cost less than a full sized standard 3 phase transformer; in fact I'd expect the weird transformer to cost more.
-Jon
