I tried briefly PackFire. In large model it made little shorter files than xz packer. Looking into output file, and depacker src. I'm pretty sure that PackFire uses LZMA method too

It is good to have src for Atari depacker. I will deal with it little - obviously it will be shorter than 2KB . Will see depacking speed - will be not fast - there are multiplications for instance, as there are in depacker I used.
1. It needs 16KB stack space, where dictionary is placed too
2. You serve dest address in one address register of CPU before calling depacker, so it is absolute. Don't see for what relative addressing would be good.
Especially as it must deal with larger files than 64KB.
3. I don't know details - see 1. It worked well with over 50 files I packed for 23 games on 1 floppy project. Longest was over 300 KB,
4. Instead time the speed: about 4-6 KB/sec .
I will test PackFire depacking speeds.
I don't think that we even should mention ICE in same thread with LZMA

It is slow, and ratio is nothing impressive.
Did depacking test. Large model depacking code len. is 700 bytes, and is PC-relative - and that's is what is useful. Speed seems some 10-20% better than UPX depacker speed. So, definitely better implementation of LZMA for Atari, 68000

Thanx for pointing on it.
Famous Schrodinger's cat hypothetical experiment says that cat is dead or alive until we open box and see condition of poor animal, which deserved better logic. Cat is always in some certain state - regardless from is observer able or not to see what the state is.