In this paper, we propose a distributed and anticipative model for collision detection and propose a lead for distributed collision handling, two key components of physically-based simulations of virtual environments. This model is designed to improve the scalability of interactive deterministic simulations on distributed systems such as PC clusters. Our main contribution consists of loosening synchronism constraints in the collision detection and response pipeline to allow the simulation to run in a decentralized, distributed fashion. To do so, we setup a spatial subdivision grid, and assign a subset of the simulation space to each processor, made of contiguous cells from this grid. These processors synchronize only with their direct neighbors in the grid, and only when an object moves from one's area to another. We rely on the rarity of such synchronizations to allow anticipative computing that will also work towards improving scalability. When synchronizations occur, we propose an arrangement of collision checks and rollback algorithms that help reduce the processing cost of synchronized areas' bodies. We show potential for distributed load balancing strategies based on the exchange of grid cells, and explain how anticipative computing may, in cases of short computational peaks, improve user experience by avoiding frame-rate drop-downs.