Depending on the power efficiency of a Bulldozer module, perhaps Bulldozer's cores could eventually filter down into AMD's low power chips. How well would a 1 module, low clocked, L3 cacheless Bulldozer perform, and what kind of power efficiency would it have? I'd be surprised if AMD didn't try to achieve maximal efficiency/mm^2 with Bulldozer.
Bocat is synthesizable, meaning it can easily be portet to new processes, without time and cost consuming manual optimations. The drawback is lack of speed. In your formula it is less efficiency/mm2. But you can have have it faster to the market, and more derivatives for less cost - very fast.
If we take BD, and hopefully some more technical knowledgeable can sheed light on this, i would guess that it is comparable larger to have better modularity. If we look at the die, there is lot of area that seem to be unused. I guess that is to make it cheaper and faster to have derivatives, (different number of modules, APU configurations, future south bridge whatever). So less efficiency again for mm2.
For both bobcat and BD AMD seems to sacrifice efficiendy/mm2 for time to market and cost for modularity.
From a strategical perspective AMD is trying to compensate for its lack of ressources compared to Intel, paying perhaps 15% efficiendy, to be able to reach the same product portfolio, with 10% of the cost. Its a solution as described by idontcares project management figure.
As for low power module BD i dont think so, the front end is very beefy, and having it running in idle must cost a lot. But i guess you are right we will se very small 30mm2, 1 module sans L3 BD ! plus gpu perhaps 80 shaders - it will rock for mm2 on the low cost market. Having the ooo running on bobcat is also costly for power. Perhaps the future very low power, will have some completely different solutions to that dilemma.
Excactly therefore knowing the Hyderabad team is there, is good to know. They seem to act very fast, and with good energy.