I have a desktop pc with a Gigabyte Z97X-SOC Force motherboard, and it has a Qualcomm Atheros Killer E2201 nic onboard.

I know the mac address is supposed to be locked in during manufacture, but for some reason mine is changeable. If I set a 'hwaddress' value in '/etc/network/interfaces', it is stored in the BIOS next time I boot. The folks at Gigabyte insisted this wasn't possible until I sent them BIOS info screen dumps showing different mac addresses, then they just admitted they had no idea how it was happening and gave up.

It did cause my Windows 7 Home Premium installation to choke and insist it needed to be reactivated every time the mac address changed, resulting in a weekly routine of phoning Microsoft for another activation code. Eventually I installed Windows 7 Ultimate and ran it without activation

For example, the original mac address is '74:d4:35:ea:07:41', so on one of my Linux installations '/etc/network/interfaces' is:
On a different Linux installation with a different hostname where I want a different ip address from my router's dhcp, I have generated a new mac address using an md5 algorithm based on the hostname and interface name:
I've attached two screenshots from the BIOS info display showing the different mac addresses.

I'm not particularly concerned about it, I think it's pretty cool to be able to do this, I'm just curious about how this happens. How does a change to the mac address in '/etc/network/interfaces' get into the nic's firmware?

I'll take a wild guess and say that that nic is a (loosely speaking) software solution. As such the traditional assumptions of bios interaction will not apply. I'll further assume that the driver interacts with the OS and has some write ability to the software. The bios reads the software much like it would do in a hardware burned nic.

I've noticed that my router (on which I installed a *nix-based firmware) has a setting that lets me arbitrarily specify an arbitrary MAC address to be used for network actions.

Kind of the same idea I guess but maybe not the same mechanism. Things like clone mac on wan port or set is common on routers.

That's probably not far wrong, although from the response of people at Gigabyte I got the impression it wasn't supposed to do that. I did once see an oblique reference to some nic firmware that had a fallback function to get a mac address from the network when the hardware mac was corrupted/damaged/inaccessible. Maybe it's something like that, but I would have thought the Gigabyte techs would know what their own hardware was capable of. They seemed to be shocked about the situation

"I would have thought the Gigabyte techs would know what their own hardware"

Call a help desk but we call it the no help desk. Very common in support.