defxor more than gets it.

Those who do not, perform this thought experiment.  Assume the price of bitcoin does NOT increase, but in fact decreases to around $1.  Now think about what difficulty will do with the current GPU price to produce a coin of around $3.

You will see that while FPGA mining makes little sense at the current price and difficulty it is inevitable if price falls low enough.  There plenty of reason to expect diffiuculty to follow price, so don't worry about relative performance vs GPU.  As the profit from GPU mining goes negative mining hardware will have to become more efficient.  In spite of lower performance it will be the only practical way to obtain a piece of 7200 bitcoins a day.  It doesn't matter if GPUs will produce 10x or 100x coins at a 50% loss per coin -- that will only be practical for the "free power" miners, and "free power" doesn't scale.

ROI is still TBD.  The current generation of dedicated mining tech is not price competitive -- but if you look closely you can see promise of pricing equivalent to current GPUs.  The first outfit to hit a $100 for 100 mhash price point at ~10 watts will make for millions in sales.  77 boards sold at $610 for < 200 mhash at 15 watts.  How many do you think would have sold at $300 for 200 mhash?   How about $200 for 200 mhash?  How about $100 for 200 mhash?  Remember that there's no longer a requirement for an expensive gaming board or PCIe extenders or industrial sized power supplies, other than a cheap recycled laptop it's all about the dedicated hardware.

There's a good reason we all expect price deflation for technology.   Saying it'll never happen is placing just such a bet.