"One question i have is what the latency would be like for games?"
Depends on whether it is Grouping/Graphing or PeerChannel. Grouping/Graphing use the replicated database model, which lends itself better to persistent data. This works for some games, but not for others. i.e. it is great for state data for turn-based strategy
games, but poor for bullet position data for a FPS.
PeerChannel is built on a more socket-like model without the persistence. This model is better for things like the bullets.
Regarding latency overall, the latency is definately higher than in a server based model (1 hop vs lg(n) hops), but it is not prohibitive in most cases. Most scenarios involving a limited number of players (say, 32) do not involve very many hops (4 or fewer),
so it is not going to be bad. Of course, it is going to be a lot better if those 32 are all on one LAN, and it is going to be worse if the players are scattered all over the globe.
Yes, zeroconf includes mDNS as part of its protocol suite. mDNS, like many other protocols (LLNMR, NetBIOS, PNRP, WSD, PNM, SSDP) can perform configurationless name resolution on a subnet. PNM differs from the others in that it is intended for finding
users over the subnet, not just arbitrary names or machine names. This is a construct more meaningful for many applications than the other protocols mentioned before. To minimize the number of actual wire-level protocols, PNM was designed to layer on top
of an existing subnet protocol, WSD.
What zeroconf does not provide (at least according to any of the docs I have ever seen) is Internet-scale name resolution. Its capabilities are targeted at the subnet. PNRP was designed to provide serverless name resolution across the entire Internet.
The fact that the protocol also allows for resolution within the subnet is an added bonus, allowing one protocol to be used by an application regardless of the environment. Furthermore, PNRP provides a rich security model that is lacking in other deployed
name resolution systems.
Regarding MS technologies "Causing problems with Zeroconf", that should not occur. There is, in fact, no interaction between the cited protocols.