I speak as a miner who has been making decisions (mining/buying/holding/selling) since when the whole network was ~100 Ghash/s... and I have told you this before, but you grossly underestimate how fast the network is going to grow in the next few months and years. "2x per year", yeah right. For reference, the network's mining speed was growing by 100x every ~6 months when GPUs were introduced. Yes. Not 2x, but 100x.
(I also think you overestimate BFL's ability to ship in merely 4 months IMHO). Because of that I don't think that anybody can assert with good confidence that one of the options is definitively more profitable than the other.
CPU to GPU was also when bitcoin started to get noticed by the entire community, so it's not really a fair comparison. I think the estimates of a 10x to 40x increase in hashrate from the first BFL shipment are more likely, and maybe doubling per year after that. We're just going to double when Avalon actually arrives. ASICMiner is going to do something similar to that.
unknowns.
You are in for a big surprise... BFL's first 100k chips are only the tip of the iceberg. A low steady increase of 2x per year after BFL ships this amount would be plausible (1) if the BTC exchange rate stayed mostly constant; (2) if competition between ASIC vendors had stabilized; (3) if there was a point in the near future where we could imagine demand stop outstripping supply; and (4) if ASICs had reached the current 28-32nm process node with no more obvious room for improvement.
But none of these points are true. The exchange rate increased 5x-6x year over year from March 2011, to March 2012, to March 2013. There are only 3-4 ASIC designers in a vastly under-supplied market. The most advanced ASICs (BFL) are only 65nm.