That DS2438 does look nice and I've earmarked it for future use, but as Bob says I don't think it helps a huge lot here although it may do if you can organize and isolated 1-wire bus.
1-wire is bi-directional which makes it hard to isolate and I don't remember seeing any isolation chips or app notes. However if it can easily be done then maybe this chip would be a good choice.
If isolation is not practical you get back to having a local processor in which case all the DS chip does is give you a temp and voltage sensors on one package.
As the Attiny85 can do both temp and voltage anyway with just a couple of external resistors (for the voltage divider) I guess it's a done deal that an extra chip is not needed.
What is needed is a protocol and there have been many options mooted, all of which will work and all of which have strong and weak points.
As for the address enumeration, how about this. All fresh nodes have address 0, they are added one at a time and the master allocates addresses starting at 1, the nodes store their address in EEPROM.
The master polls the nodes (as well as address 0) using one of the protocols we've talked about.
If a node dies it is replaced, the master detects a) that node N has disappeared and b) then node 0 has appeared. Being that the master has some really clever code written by someone of equal or greater intellect than the contributors to this thread it puts 2 and 2 together and allocates address N to the new node.
This way you can have multidrop to "fix" the problem of one dead node killing the entire chain and have auto address enumeration as well.
FWIW I reckon the daisy-chain topology is potentially more reliable than multi-drop, but only if you have redundancy in the form of two chains working in opposite directions. Of course I have to believe that because that's exactly how I'm doing my current project