To realize the benefits of network connectivity in transfer-based transit networks, it is critical to minimize transfer disutility for passengers by synchronizing timetables of intersecting routes. We propose a mixed-integer linear programming timetable synchronization model that incorporates new features, such as dwell time determination and vehicle capacity limit consideration, which have been largely overlooked in the literature on transfer optimization and timetabling problems at the scheduling stage. We introduce a new concept of pre-planned holding time, called transfer buffer time, to reduce the transfer waiting time, particularly for transfers to low-frequency routes, while taking into account the penalty of extra in-vehicle time for onboard passengers and the possible consequences on headway regularity of a route. We develop a Lagrangian relaxation-based heuristic to obtain high-quality solutions efficiently in applications of large instances. Our experiments on instances with up to 12 transfer nodes in the City of Toronto, with a mixture of low- and high-frequency routes, illustrate the potential benefits of the proposed model over a conventional model representing the state of the art found in the literature. The results indicate that incorporating transfer buffer time, dwell time determination, and vehicle capacity limit consideration improves model outcomes considerably. The experiments also demonstrate the computational efficiency of our Lagrangian-based solution method compared to a commercial solver.