The question should be when is software going to fully utilise dual/quad core. Not when is oct core due out.
Software can fully utilise as many threads as it NEEDS now.
People get hung up on this, theres a big deal about how Crysis doesn't really use quad cores. Except, it does, it spreads its load quite well, will happily run different parts in 4 threads. but it barely puts a 40% load on my computer, why, it simply doesn't NEED any more power.
For some insane reason people were harking on about how Crysis should for some reason put a 100% load on all 4 cores for no real reason.
If you want to render a 3dimage you can get programs that will use all the cores at maximum. however, you'll be hard pressed to find a web browser that uses all 4 cores to the max, but it simply doesn't NEED to.
All ported games from PS3/360 to the PC use all 4 cores(well most at least) yet none put the cpu under full load, because gaming doesn't require 4 3Ghz cores. 8 cores won't change that.
Basically all applications that NEED multithreading and fully use a CPU, can and do use 4 cores. Almost all other applications simply don't need that level of power, its a ridiculous waste of power to make firefox, IE, most other applications multithreaded when they can't even load one core.
Nehalem, as its been touted so far, isn't offering anything great performance upgrade wise. Bus upgrades, memory controller ondie, its all sidesteps. Instead of 12mb cache and lots of on die logic to keep latency down, you replace it, key word REPLACE< not add to that logic, completely replace that on core logic. Its a sidestep to future technologies, it should barely if at all improve latency, without decreased latency, theres no performance boost to be had. Infact, in situations where the smaller cache miss predicts and it accesses memory it might be slower in some things, random stuff, like games, maybe not, ordered things, IE scientific apps, 3d rendering, encoding, massive cache and great prediction can be better.
Either way its a move to future stuff, like pci-e connections on cpu die, both Intel and AMD are moving that way on future chips, you would have to have mem controller and a wider bus to do that in future. Also, multi cores with an intergrated gfx core on the cpu die, again bus and controller would be required on die to make that happen. A mem controller isn't the be all and end all of performance, its just a controller, it doesn't make a huge difference.
AS i've said in other threads, mem controller might be required for that stuff later, right now it offers us, the end user, next to nothing. Except possible yeild, clocking, power, heat, timing and other problems. If it doesn't overclock to 500mhz bus stable, then a new chipset, or a nvidia board, or a SIS board won't help, a new chip is the only thing that will fix it in the future. The X2 controller was fine, the 754/939 controllers were great, their phenom controller, not so good so far, or does something that mobo makers are having issues with.