This is another test of night time HDR with image stacking:

More details on how I did it can be found here
I am sorry to be so late here; I found this post by googling.
I think I know a way you can process your data to get better results.
Your "base" data is 256 x 2 second exposures.
I believe that these could be processed repeatedly, with ImageStack, varying all the way from "average" to full "sum" by altering the parameters.
e.g. varying the divisor from 2,4,8,16,32,65,128
This should give a sequence where each tonal value is "well exposed" in one of the processed images.
You can then HDR stack these exposures in enfuse (or similar).
http://wiki.panotools.org/Enfuse BugBear