Apparently the jittering that occurs between when the buffer is filled and the text is printed has to do with the now arbitary amount of time my buffering routine runs for, verse the original fixed "one byte read, one byte written" code. I don't know very much, but based on what I'm reading in Jeremy Chadwick's Nintendo Entertainment System Documentation, I'm thinking maybe I need to keep track of VBlank status to make sure I'm only reading from (writing to?) $2007 during a VBlank. This document is great.

This is fun stuff.

Ps. This is elite. Go Memblers.

Alright, thanks to Zooka and joat of #nesdev on EFnet, I've learned that my routine has to be small enough to fit into a VBlank period, which it doesn't even come close to doing right now. Time to figure out how many CPU cycles I have to waste, then try to see if I can cut the buffer size in half in order to gain some more.

Updated: So apparently vertical blank time is "precious" and should pretty much only be used to communicate with the PPU, which for some reason is where the original text loading routine is. Anyways, the lesson is that this is not the place to implement my decompression scheme, and instead I'll have to implement it somewhere else. At least we're getting closer to some closure here.

And still, my relationship with the NES is slowly growing up. :)