From what I understand, the DS encryption generates some data tables from the gamecode, then uses them to encrypt/decrypt commands. This data is what's in the rom at 0x1600 and 0x1c00.

You can see the details at gbatek key1 encryption details. This data is called "keybuf" there.

I can think of 2 reasons they did it this way.

1: efficiency. This way the card doesn't need to calculate the table, and doesn't need temporal ram to store it. Why require extra ram if the data never changes, and you have plenty of rom storage already?

2: more security: By doing it this way, deva never needed to encrypt/decrypt anything in their machines, so Nintendo's tools didn't need to have code for that, protecting it from getting reverse engineered.

Nintendo would assign the gamecode to the game, then give the deva the encryption data for that gamecode and the first 0x800 bytes of the secure area, already encrypted for the gamecode too (what they call the "syscall library").

Also interesting is that doing it this way didn't give devs info useful to pirate/dump other games than their own.