A few other points:
We have found that the gold-tin intermetallics that form when the gold plating dissolves into the bulk solder will migrate to the nickel plating.  They seek out the nickel, which lowers the energy of the compound when it dissolves into the gold-tin, and since nickel doesn't move very fast into the bulk, the gold-tin moves to the interface and you'll have (Au, Ni)Sn4.  This gold-tin layer itself is what's brittle, not so much the gaps left behind from when it moved.  At least that's what we saw.  We also saw that when there is copper on one side of the joint (no Ni plating), the AuSn4 didn't move into a continuous brittle layer.  The AuSn4 became (Au, Cu, Ni)Sn4 --  the theory is that the Cu is taken in where Ni would have gone.  Cu migrates faster, so is more available in the bulk.  So, you can get away with higher Au content when you don't have nickel plating on both sides of the joint.  But not too much more Au.  You should still be worried above 5 wt%

What I want to know is if anyone has successfully removed gold from a QFN without the ground pad being too lumpy to solder or the part being fried from excessive heat.