This paper describes the architecture and FPGA-based implementation of a massively parallel processing system (IANUS), carefully tailored to the computing requirements of a class of simulation problems relevant in statistical physics. We first discuss the system architecture in general and then focus on the configuration of the system for Monte Carlo simulation of spin-glass systems. This is the first large-scale application of the machine, on which IANUS achieves impressive performance. Our architecture uses large-scale on chip parallelism (≃1000 computing cores on each processor) so it is a relevant example in the quickly expanding field of many-core architectures.