We propose a peer-to-peer event ordering and simulation technique aimed at networked real-time action games. Partially based on replicated simulators, its goal is to support decentralized playout in small-scale game sessions on instanced action spaces while being resistant to collusion cheating. The action spaces are linked to persistent-state social spaces of larger scale which are supported by centralized simulation. Together, these two kinds of spaces offer support for massively multiplayer on-line games (MMOGs) that offer a mix of socialization on large-scale persistent environments and fast interaction on small-scale temporary ones. Although player nodes on action spaces are required to run a conservative and an optimistic simulator simultaneously, we show that 2.2 simultaneous simulation steps are executed on average and that 11.95 simultaneous steps occur as the average peak situation for 20-player sessions with 150ms to 300ms network delays between nodes, 5% probability of any late events introducing errors, and rollback and re-execution operations having their execution spreaded through 100ms of real time or longer.