The most logical answer would be that miniaturization takes infrastructure to create, build, design, power and maintain. The future even at its most technologically advanced did not appear to be a place where advance scientific research into micro-miniaturization is taking place. That does not mean it isn't, only that we have not seen it. But it seemed quite a stretch given the difficulties they had just keeping their temporal mechanics technology running.
Their greatest technical limitation is likely a power supply for their technology. When you look at what Skynet creates, it is almost always making things that are massive. Miniaturization, as we are learning today takes time, energy and creativity. While Skynet has time and energy enough to power its larger devices, it lacks the creativity to make smaller, effective devices that could exterminate mankind. And when you are talking about nanomachines, Skynet has proven it still has quite a ways to go. Yes, it does have the T1000 but their limitations, and there must be some, otherwise there would be no reason to create anything but T1000s. Those limitations likely include, production facilities to create them in, raw materials to make them from, programming technology for their interface and programming design and powerplant design (what do they run on anyway?).
We are told the power sources for the Terminators are rather large, bulky and by our standards reasonably effective. But I suspect they don't scale well. Whatever materials are used to create the Terminators requires huge infrastructure to acquire, refine and create the power sources used by the machines; certainly from an infrastructure perspective quite costly.
We have discussed on the Scifi.Stackexchange, the question of why all of the metalloid Terminators only seem to create single physical structures. It was theorized by most that the machines did not have the capacity to make individual nanomachines due to power constraints.
There seemed to be a limit to how intelligent or effective the nanomachines were after they reached a certain size and that larger was better.
To be honest, if Skynet's goal was the destruction of humanity, it is certainly mucking it up. There are dozens of ways it could be more effective than it has been.
Why make Terminator sized robots at all? Create robots the size of cats or dogs, equip them with razor sharp claws and teeth. Find, follow and track the rebels to their bases. Build ten cats for every Arnold sized terminator, penetrate the facilities through holes and vulnerabilities in their strong holds and physically rip out their throats.
Why not create a nano-probe mist? It could be air dropped with nanites that replicate upon touching human flesh, feeding off the organic energy for a time. Otherwise they just sit and wait for human warmth to trigger them.
They are then brought back to the human strong holds and after spreading for ten days, they simply strip away enough calcium from the bones or iron from the blood, or cause aneurysms in their brains.
Why didn't Skynet do any of these things? Skynet doesn't seem to have much imagination. It seems to use the data, (since it appears to have first appeared as a military program) designed into it to use standard confrontation tactics. It later learned espionage and guerilla warfare from the rebels. Skynet appears to learn, but only very slowly. If Skynet were a truly thinking computer, and its goals were to exterminate mankind, we could safely assume, humanity would be dust already.
The TV spin-off series implies that Skynet is of two minds and perhaps the complete extermination of humanity is not a completely done deal. Perhaps this is why the time-travel jaunts have been so effective at retarding its technological capacities, keeping it from growing more sophisticated and having to spend more time and energy just holding its ground.