Hi wd8ivl,
I agree with all prior posting re: the Q as posted.
The attached zip has a LV 6.1 example along with a jpg illustrating two methods that could be used if we take a step back and loosen some restrictions.
Both methods build the cluster using "bundle by name".
The first just insert the sub-sets of data into a 12 element cluster in two steps. This has the drawback of possibly dragging unrelated data around when it is not needed.
The second uses type defs that avoid the "extra data copies" and picks-up the luxury of the sub-VI used to combind the data will automatically adapt when the cluster definition changes. This method does not give you a flat cluster in the end. It also requires you have a unique sub-VI for each cluster type.
Like I said, it may not with what you are attempting, but these methods could be helpful for similar situations.
Ben