If I had an already-profitable miner I'd probably run it full-bore until it hit breakeven returns, then underclock/volt it about halfway and keep running it until breakeven returns again, and then take it down to the bottom. That's just what I'd do, I haven't checked the numbers to see if it'd be better or not than running it at bottom speed the whole time.
A lot of chips dissipate most heat through the top. ASICMiner's BE300 would have been a top cooler; so is the Rockerbox and all those other big-die chips. There's no requirement for string design to have capacitors on the bottom.
I guess I don't honestly know if it's designedly a top-cooler chip, but one of the reasons for doing that (especially the chip was designed to be string'd) is because in a string design you have multiple local ground planes all at a different absolute ground potential and if they short through your heatsink things can catch on fire if your PSU doesn't trip out first. Maybe I'll test out heatsinking on our breakout board and see if one on top or one on bottom keeps the chip cooler, but that'll only matter for the USB miner. The TypeZero would have to be top-cooled so the heatsink doesn't kill the power planes.
One good thing about building a flexible miner (and by flexible I mean with software-adjustable voltage and clock) is you can make it a high-speed miner or you can make it a super-efficient miner, whichever you want. Our board should have better low-end efficiency than the S2 upgrade's fixed operating point. Actually, two of our TypeZero boards should be just about equivalent to one S2 Upgrade board (by chip count), but ours can operate at both higher speeds and lower speeds because it's not a fixed voltage. A single board can operate like a U3 (that's why we chose our board size and USB connectivity), except you'd be seeing a similar price and quite a bit more hashpower.
There is good demand for not-crappy not-crappily-priced USB sticks. We will do our best.
Also, Guy tells me they actually scrapped out all their Hammer ASICs.
How far are you with testing heat dissipation? AFAIK most of the chips are made bottom dissipated, this is done because high power chips are usually soldered on heatpad, so heat transfer through solder balls (or pads) is better bottom. Don't know if BM1384 is this case. Also, this can be seen on A1 chips or older BM chips... Different situation is with big-die chips, they are upside-down with direct or indirect contact to heatsink.
I thought that string design needs bigger capacitors closer to Vcore pins to bypass transients caused by chips current draw variation.