The German Aerospace Center (DLR) is developing ScOSA (Scalable On-board Computing for Space Avionics) as a distributed on-board computing architecture for future space missions. The ScOSA architecture consists of commercial off-the-shelf (COTS) and radiation-tolerant nodes interconnected by a SpaceWire network. The system software provides services to enable parallel computing and system reconfiguration. This allows ScOSA to adapt to node errors and failures that COTS hardware is susceptible to in the space environment. In the ongoing ScOSA Flight Experiment project, a ScOSA system consisting of eight Xilinx Zynq systems-on-chip with dual-core ARM-based processors and a LEON3 radiation-tolerant processor is being built for launch on DLR's next CubeSat in late 2024. In this flight experiment, not only all 18 cores but also the programmable logic will be used for high performance on-board data processing. This paper presents the current hardware and software architecture of ScOSA. The scalability of ScOSA is highlighted from both hardware and software perspectives. We present benchmark results of the ScOSA system and experiments of the ScOSA system software on ESA's OPS-SAT in orbit in combination with a machine learning application for image classification.