Long-baseline time transfer can reach high frequency stability based on post-processed batch least-squares adjustment using precise point positioning (PPP) or integer-PPP methods. For real-time PPP users, time transfer results are degraded by filter-based processing and by real-time satellite orbit and clock products. Real-time kinematic (RTK) time transfer can reduce satellite-related errors but has baseline-length limitations. In this study, the PPP-RTK technique, combining advantages of PPP and RTK, is tested for real-time long-baseline time transfer. By delivering satellite clocks and satellite phase biases produced within a PPP-RTK regional network, time differences can be estimated between users and a reference network station. Using dual-frequency GPS and Galileo data on an approximately thousand-kilometer-scale network in Europe, the method is evaluated on the 884 km BRUX-ONSA and 920 km WTZR-ONSA baselines. At an averaging time of 10^5 s, modified Allan deviation at the sub-10^-15 to 10^-15 level can be reached. In kinematic mode, median clock residuals converge to 1 ns and 0.3 ns within 2 min and 15 min, respectively. With simulated observations from 150 low Earth orbit satellites, convergence reaches 1 ns and 0.3 ns within 30 s and 3.5 min for all three estimation modes.