Networked embedded systems present challenges for designers composing distributed applications with dynamic, real-time, and resilience requirements. We consider the problem of recovering from failures of distributable threads with assured timeliness in dynamic systems with overloads, and node and (permanent/transient) network failures. When a failure prevents timely execution, the thread must be terminated, requiring detecting and aborting thread orphans and delivering exceptions to the farthest, contiguous surviving thread segment for possible resumption, while optimizing system-wide timeliness. A scheduling algorithm (HUA) and two thread integrity protocols (D-TPR and W-TPR) are presented and shown to bound orphan cleanup and recovery times with bounded loss of best-effort behavior. Implementation experience using the emerging Distributed Real-Time Specification for Java (DRTSJ) demonstrates the algorithm/protocols’ effectiveness.