Yep seen that, Nick has a lot of good tutes on his site.
I originally had a multi-drop bus with a master, but changed it to the current redundant-ring-multi-master topology, I think it is potentially much more robust because even with 2 or 3 faults you still have a mostly (or even entirely) working network. Multi-drop-master only needs one small fault and the whole lot collapses in a heap. That said I guess it's been good enough for all manner of networks over the years.
In fact I had Attiny84s as well, but wanted to go faster than they could handle. At present I can get 500kbps which is getting close to what I'd like, admittedly that's not continuous throughput yet but it should work at that speed or better.
It sounds like your idea is very similar or even the same as my original design, with a very fine node granularity, ie one node one function. As you've seen I've moved away from that a bit because no matter now many things a node does it still needs the same network overhead. So I decided to amortise that overhead over multiple functions.
That said the APE chips will be capable of working stand alone, in which case we're back to a tiny 3-chip node for just a few $.
My thought was also to send the power over the the connection like power over Ethernet.
As I to have done, it's not always appropriate due to voltage drops, wire ampacity etc but when it is it saves a lot of mucking around.