Speaking of 'like the BSDs,' I'd have to argue that, from an outside perspective, AROS appears hobbled by adherence to its own design principles. That is- a lot of good thought has been put into the organization of the system as a whole, but since commits are held to that standard, Good Thought has to be put in. Meanwhile, the tree was held private for a long time, and in general, the whole process of involvement has been somewhat obfuscated by the desire to keep the project 'steered' and manageable.
The problem is, there's no easy starting point to bring code into the Amiga fold.
Now, I'm just blathering at random, but does an AROS Research OS seem too farfetched? That is, an experimental NetBSD to AROS's "Production System,*" held to the barest minimum technical requirements - commits will be accepted so long as they pass Tinderbox? With an accompanying cvsweb and other 'convenience features' for public involvement, it seems it could make a good 'playground' for projects and ideas that could then roll back to AROS ("Keepers of the API") or other systems at the discretion of those projects' maintainers?
I'm not sure what such a build would look like; it'd probably be a munge, maybe with a lot of BSD warts (the AROS license is a bit complex to wrap one's head around, but I don't *think* there's a conflict there) ... but if it grew enough warts to make it easier for external projects to build, those builds could then be that much closer to running on 'pure' AROS, OS4 or MOS...
Perhaps there'd need to be a little more restriction to ensure things diverge beneficially- maybe the distribution should stay synched with AROS exec and/or certain base libraries(?), to ensure the free-for-alled development would maintain the 'genetic' relationship to bring the above to pass... but I'm just sketching here, not in any way trying to make a full proposal. (I'll readily admit I've never had the time and resources to play with AROS, let alone flip through the source!)
A lot has happened since the project was founded... This certainly wouldn't be a magic bullet, especially after all this time, but I don't think it's deniable that more work would've happened if a miracle happened, things went GPL, and the Slashdot crowd had somehow been made to be interested. That such work wouldn't have upheld the goals of the project as well is also obvious- so the question is, "Why not have two projects?" (I know the answer is, of course, "Fine, host it," and I'm certainly not equipped for that... but maybe someone else is?)
---
*Or in this case, "Reference Implementation," given the focus on architectural purity.