When the small block chain eventually takes over the BU fork, are the BU nodes going to switch to that chain and basically erase the forked chain from memory?
It's
supposed to work that way. Not just "Core 1MB", but any miner can set any blocksize at any time, and if that miner has guessed right and enough of the network accepts it then that will be everyone's new setting. Otherwise that whole chain will be orphaned. It doesn't matter if the blocksize goes up or down. That's what "emergent consensus"
means.
Depending on what you signal, you could trick your fellow miners to mine chains that will have a low chance of survival. The more hash power and nodes you control the more you can potentially gain, but it's all subject to the usual variance to there is an element of chance to it.
Nobody really knows how that would play out, since nobody has taken it seriously enough to study it in detail and run any realistic simulations or testnets. How many confirmations will it take to protect against double spends? Nobody knows. The chain you're on can potentially be invalidated at any time. How large blocks will we end up with? Nobody knows. They'll probably get both larger and smaller.