One thought that immediately comes to mind is to break this into two models. While this will not immediately resolve your issue, it will prevent some others that you may not be aware of. One typically does not want to involve many processes that don't need to be rerun over and over with an Iterator, as every time the Iterator fires it will typically try to process everything in the model again.
Instead, break this into two models, then create a third one to link the two models. This way the first part - the processes leading up the Iterator - only run once, and will then exist for the second model to access when it Iterates.
In terms of your original question, there are a few potential solutions. I suspect the solution is one of these, but you will have to play with it a bit with your particular data.
Chris Donohue, GISP
