1) PoW will be updated as hardware grows bigger...cache.
How? Do you think that you are going to get 51% of the miners to accept such an update, once the first hardware implementation is available and any significant minority of miners have already invested into same?
Um, adoption dynamics of tenebrix are such that majority of miners are, for various reasons, riding a general purpose computation device which is either unfit for installation of custom hardware or would require some overhaul to do so
Of course they will accept something that will keep an annoying - and potentially game-changing - competitor out.
If you are a cpu-centric miner mining a decidedly cpu-centric chain, you have a vested interest of keeping as much "exotica" out as possible (I wonder tho whether APU systems are "exotica" - they are essentially beefy CPUs)
I'm not limiting hardware implimentations to GPU's and FPGA's; and the entire argo doesn't need to be implimented in order for the use of hardware to be cost effective.
ASICs are just stupidly un-economical and carry additional bonus of being (Unlike GPUs and FPGAs) pretty truly dedicated
If there is a way for it to be accelerated in hardware, someone is eventually going to figure out how to do it.
You forget economics.
There's no point of deploying an application-specific system unless the factor by which it accelerates mining makes it economically viable, and dedicated custom solutions suffer from this far more than semi-dedicated such as GPUs or FPGAs (which have apparent uses outside mining and can be, worst come to worst, re-sold on general market unlike Hypothetical Custom Tenebrix Supercruncher that can only plow through scrypt)
You could crunch TBX very fast on, I dunno, a non-uniform memory access system, but it would be prohibitively economically unsound unless you know a way to get one for free (in which case I love you and want to get to know you closer
)
The more exotic the hardware that is required to do this, the more centralized that mining becomes. GPU's are no longer uncommon hardware, even if there are still many consumer desktops still in service that don't have suitable ones. You can't really buy a modern destkop that doesn't have a CUDA capable GPU, unless you shoot for the cheap. If it's an iMac, you quite literally cannot buy a new desktop without a CUDA GPU. And even though FPGA's are expensive and somewhat exotic hardware today, their general purpose usefulness pretty much garrantees that many consumer desktops are going to have intergrated FPGA's on the mainboard in another ten years.
Ten years down the line, I fully expect TBX to undergo no less than two major PoW updates (assuming it survives)
However, if you succeed in making it uneconomical to mine with these forms of commodity hardware, only the truely exotic hardware will be able to do so, and you will become dependent upon the purchase of PCI cards with such exotic hardware in order to expect to compete cost effectively for mining once major operations start using these purpose made devices.
Again, this assumes that economics of producing totally-custom TBX-crunchers render them competitive economically against CPUs which, through benefit of being among most common and mass-produced computational systems, have vast economies of scale driving their
price down.
Doesn't matter how good your million-dollar supercooled Tenebrix-cruncher is, if I can achieve same performance at a fraction of the cost by just "buying moar CPU-centric boxen", you will lose.
Throwing monkey-wrenches into various hardware optimization attempts seems like a fun field, perhaps a science unto itself.
P.S.:
Not to mention the opportunistic-mining implications of having an algo specifically tailored to be most economically sound on the most common computation substrate available
