When the program loads a snapshot file, it scans the dumps for sysex headers and looks at these to figure out what type of dump it is. It then loads the dump in the corresponding slot in the snapshot. This means that it doesn't matter what program created the sysex file, nor does it matter what order the dumps appear in the file. The program creating the file only needs to write ordinary TX81Z bulk dumps to the file in order to be loaded directly into the snapshot of the TX81Z Programmer.
When you are creating a snapshot to be saved to a file, it doesn't need to contain a complete set of dumps. You can create a snapshot of just one bank of voices if you want.
If you have a file with multiple dumps of the same type, only the first one will be loaded, since the snapshot only has room for one. The only exception is with the voice banks. Since there are five of them, it will load up to five voice banks into the snapshot. The first one will go into bank I, and second one will go into bank A, the third bank B, etc. The program doesn't try to figure out which banks are preset banks or anything, it just loads them into the slots. You can't transmit the voices in the present slots to the unit, anyway. They are just there so you can manipulate them. If there are more than five voice banks in the file, the rest will be ignored.