Hi,

Would an AC servo driver be able to respond that powerfully that fast?
Yes, that is what bandwidth means. Its not new either, DC servos of yesteryear also did well, maybe not 5kHz, at least not cheaply....but that is what bandwidth means.
It is a measure of the time taken for the system to respond to a disturbance.

I am a hobbyists and I have five 750W Delta servos on my machine, or until I finish my fifth axis, four servos and one waiting to be fitted. The rated torque for a 750W Delta servo is 2.4Nm,
and overload torque is 7.1Nm. I would expect any one or more servos to go from idling along at 0.5Nm to 7Nm in response to a sudden increase in load within two time constants, with the time
constant being the inverse of the torque loop bandwidth, ie 5kHz. Thus the time constant is 200us.

So I would expect any of my servos to go from idling along to max output in 400us....which I think you'll agree is pretty damn quick. I would expect no better, nor even as good, a response
from a realtime computing platform.

These servos cost me $438USD each....so not cheap exactly but so much performance for the sum expended.

The microcontrollers inside a servo drive are very impressive. I'm using a Texas Instruments TMS320F2069M micro controller for a servo drive I'm making. These things are bloody great,
single cycle 32bit floating point multiply and accumulate, nine highly configurable PWM peripherals, in fact peripherals up the Yazoo, running at 90Mhz, and all for under $20.00USD.
These things are just made for polyphase rotating machinery and I don't believe there is a general purpose computing platform that can close the loop that comes within a bulls roar of these things.
It is controllers like this that have made modern servos as good as they are...back in the nineties they would have been so expensive, only the military need apply, but now they're everywhere.
Even the cheapest crappyist of Chinese servos has one of these controllers inside that can close a feedback loop with vastly greater precision and speed than we can imagine, and all for less than a
'feed of fish and chips'.

Maybe the key words here are 'more reliable'?
But even so, a dedicated (FPGA?) controller is very likely to be better.
For my money I'd rely on one of these purpose-built microcontrollers before I relied on any computing platform....mind you that because I've programmed them and de-bugged code to run
a Field Oriented Control algorithm and can personally attest to just how potent they are. If you've not been up close and personal with one you might be forgiven for not realizing what a game
changer they are.

It is that lack of knowledge that means that many believe that 'full close loop control' must be done by a realtime computer like LinuxCNC or a reatime controller like Galil when in fact a humble,
cheap, 32bit microcontroller can 'hold them to the flames', at least for rotating polyphase machinery.

Craig