@ggn: Thanks for the heads-up vis-a-vis the GemDOS header; I never would have guessed that's what it was. The header is $20 bytes long in this case, and the load/run address is at byte $1C in little endian. So the load/run address for these demos is at $460.
The size of the payload is at byte $02 in big endian. What a strange format. I wonder what tool it was that was used to make them...
Nope, the $6004 you saw is actually a bra.l *+4, which simply skips a long and jumps into the main code. Now, the long that is skipped might look like the load/run address, but my disassemler (Easyrider) can't find any reference to it. Further down the code though the address to the bra.l is fetched and the exact amount of bytes that consist the binary is copied to $6000 and then the code does a jsr at the address in that long ($8720). Now, I really don't think that jagserver would go into the trouble of adding such a complex overhead (especially since it has quite a chunk of init code). The weird thing is that the code seems to be position independent (using PC relative everywhere and all) so at first glance it would appear that it could run even when uploaded to RAM at an address larger than $6000 but it doesn't (of course I could be of help more here if some emulator had a built-in debugger, right? )
There you go Shamus, I tried to spare the world from all this babble but you forced me to put everyone to sleep - happy now?
But it didn't appear that they were done by the same artist
HOL and Mobygames say it's the same. After all, since Jochen spent lots of time making a cross-platform format (TFMX) for writing music, it'd be silly to let other people do the music