As data-intensive applications proliferate, cloud computing is transitioning from single-cloud deployments to collaborative JointCloud paradigms. In such cross-cloud environments, a complex interdependence between coalition formation and multi-dimensional continuous resource allocation is induced by resource heterogeneity, price and QoS disparities, and decomposable yet volatile user demands. Meanwhile, the collaboration structure determines resource accessibility, while the allocation outcomes affect system efficiency and user experience, creating a tightly related decision process. Existing optimization, matching, or market-based mechanisms often fail to jointly address this relationship, leading to suboptimal suitability, fairness, and adaptability. To address these challenges, we propose a mechanism framework called M3A-JC (Multi-Dimensional Multi-Party Matching Allocation Mechanism for JointCloud), which combines many-to-many stable matching with multi-dimensional resource allocation under quota and capacity constraints. In M3A-JC, bilateral preferences are constructed through normalized QoS suitability, price acceptability, demand–capacity compatibility, and a CSP-side cost-based admission score. A many-to-many deferred acceptance process then obtains a quota-feasible pairwise-stable matching. A resource-dimension-specific proportional rule allocates continuous resources over the established associations, with capacity perturbation handling and utilization control accounting for bounded capacity variations. The analysis establishes finite termination and pairwise stability for fixed score-induced preferences, together with feasible deterministic allocation under fixed realized capacities. Path delay, bottleneck bandwidth, and completion deadlines determine eligible user–CSP associations, and deadline-aware CPU allocation reserves the minimum feasible share for each admitted association. Repeated-seed experiments evaluate welfare, fairness, sensitivity, runtime, optimality gaps, strategic reporting, topology-constrained deadlines, workload variation, and rolling-horizon rematching. At 50% joint α–β manipulation, the mean true-welfare loss is 1.60%. The topology tests record zero deadline violations among admitted associations, and the Alibaba trace-driven profile gives 3.66% higher mean NTSW than Elastic-Orch. Incremental rematching gives a −0.099% mean NTSW difference from full rematching (p=0.458) and reduces migration by 72.05%.
