1) There was large scale mishandling of the initial stages that made it look like we were just going to jettison code that wouldn't be relicensed or force people into things they didn't like. Contrary to some external appearances, MAMEdev is not all that organized, and the "48 hour rule" always applies (wait 48 hours if you don't like an apparent decision, it'll change).
2) The ownership details are being handled according to relevant US and European law, and a majority of them have come from Balrog, not Aaron. The upshot is that if you change functionality and it's not a single-liner you generally get some ownership. Mechanical reformatting of a file (e.g. the modernization we've seen a lot of lately) doesn't grant ownership because the actual functionality didn't change and remains how the original author(s) did things.
3) Going down his specific examples: Stephh gets part-ownership for DIPs. Mash/Mamesick hacks that are more than a single line get part-ownership. Refactoring and modernization don't because the functionality of the code doesn't change. Adding save states gets part-ownership. EEPROM/sound hookups get part-ownership in the case where it's not a single-line AM_RANGE() thing. (And I've done a lot of those in both forms). Decryption functions get part-ownership. Tilemap conversions are a gray area; since you're changing the actual functionality of a great deal of code (even if the results are the same) I'm inclined to give part-ownership for that.
4) The authors of any research or manuals you use to write the code don't own your code; Motorola/Freescale doesn't own the 68000 core, and Charles MacDonald doesn't own anything he derived information for unless he actually wrote the code. That said, you're an asshole if you don't give Charles effusive thanks in the comments and a link to his website. Ditto Guru readmes, which generally are self-crediting.