Memory interleaving allows you to avoid a bottleneck communicating to the RAM module. If you're trying to push data to the chip faster than it can support it, you can send some data to its partner instead. Now if you assume that you write to both chips simultaneously, almost in a one-for-you, one-for-me fashion, you see why the chips need to be laid out in a symmetrical manner. RAID hard drives, then optimised for speed, work on a roughly similar concept.
As for your situation, if you really only want 16GB, you could just go for 2 more 4s. But if you want to put 2 8s in there, that will work fine too. That's what I have (24GB, 4-4-8-8) and it works a treat.