The 128K works differently than the 512K. It only loads into RAM whatever it needs from the System to do what is requested at any given moment, and then immediately purged. That's another reason why it's so slow. It's very efficient, but unlike the 512K does not load in a minimal static system into RAM. This is also the reason why it will not do AppleTalk, because it purges the polling scripts from RAM and therefore would go off the network every time you requested a different task from the Mac. Rather than writing custom code for the 128K, they just excluded it.
I recall reading that there is a minimum system heap loaded into RAM, but its much smaller than the 512K. I thought this in for was included in Inside Apple, but I can't immediately locate it. Perhaps someone here will know where it can be found. This is why there is such a discrepancy between the 512K System heap and the total RAM available on the 128K.