The physics in PewPew appear to be deterministic. Rather, they appear to be pseudo-deterministic. Two clients running on identical processors connected over the net to a similar server (without any dropped events) either will or will not diverge (this is apparently determined at the time the players join). If they will, this happens pretty quickly, and can be exacerbated by a rapid number of collisions. If they will not, then arbitrarily fine maneuvers can be replicated deterministically between the two clients.
This likelihood of one or more clients diverging increases with the number of connected clients. So, two clients will nearly always stay in sync, three will rarely stay in sync, and four will almost never. In these situations, it is not uncommon to find several pairs of clients which each share a unique state. Thus, the possible number of diverging states is probably discreet (since some clients become identically wrong, never nearly identically wrong, nor similarly wrong).
The above would indicate a mis-ordering of some list internal to the physics engine (Box2D). Any advice on this issue would be greatly appreciated.