> if its simple enough for an arduino, but you need the OS / USB / ethernet of the Pi, just skip the arduino and use a Pi by itself.
Which is the same as saying:
> if one part simple enough for a microcontroller, but you need the OS / USB / ethernet of a computer, just skip the microcontroller and use a PC by itself.
The two two are not identical. The difference is immediately obvious to a professional embedded system engineer. Cost. For an engineer, cost is the second and third most important factors. Second is unit cost, and third is development cost. Replacing a Pi with a PC fails the unit cost test, and using an arduino and Pi together when only one is necessary fails both tests. The key difference between an amateur designer and a professional engineer, is that typically the amateur is on a constrained budget which dictates that development cost is paramount. For an engineer, the equation is a lot more complex, which is why an engineer can justify spending $1,500 for an anual Orcad license, and an amateur can't.
Which means no add-in cards or USB peripherals, ever. After all, your Core i5 COULD be used to bit-bang ethernet. Therefore it should, right - no use case for an ethernet controller.
Full speed ethernet can't be bit-banged. There are too many timing related issues that would make a bit-banged solution unreliable. Audio on the other hand can and is done by bit banging with a little DMA. That is why CPU based "sound cards" are a thing. Ethernet and USB by contrast use different voltages from CPU core voltage requiring external hardware anyway. In modern PCs, this hardware also handles the protocol, so there is no cost advantage to using a phy only chipset and bit-banging USB.
As another example, the Pi can connect directly to a wifi chip, so it would be stupid use a separate wifi module like the wipi, right? You'd never do that, because you COULD skip module and integrate everything onto one board.
Depends on the unit cost vs engineering cost. If the total cost of development of the chip version is $X, and the unit cost difference $Y times the total estimated unit sales Z is less than $X, then there is no point in using the chip version, the USB wifi has the greater ROI, and should be selected.
I could continue, but apparently /. doesn't like long posts anymore...