Encoding a single image with the CPU takes nothing compared to modern video codecs.

Burst mode captures 10 images per second, for encode demands of 120 MP/s. That’s half the throughput of 4k30.

Is there a cool down between bursts or can you do 10 a second for as many consecutive seconds as you like? You only have to get the images encoded and stored before the next burst starts, or at least before the user runs out of memory.

I’ve done bursts that lasted for 2 minutes.

There's literally no way Apple developed a hardware accelerator for JXL encoding. So they're definitely using some kind of a CPU or generic DSP.

They might have some things like DCTs and colorspace transforms accelerated.