Advances in automotive industry as well as computing technology are making autonomous vehicle (AV) an appealing means of transportation. Vehicles are beyond the traditional source of commute, and leveled up to smart devices with computing capability. As one of the compelling features of AVs, the autonomous valet parking (AVP) allows for navigating and parking the car automatically without human interventions. Within this realm, long-range AVP (LAVP) extends auto-parking to a much larger scale compared to its short-range counterparts. It is worth noting that AV mobility is a pivotal concern with LAVP, involving dynamic patterns related to spatial-temporal features, such as varied parking and drop-off (or pick-up) demands with diverse customer journey planning. We herein target such critical decision-making on where to park and where to drop/pick-up upon customer requirements during their journeys. Concerning in practice that car parks are equipped with limited parking space, we thus introduce parking reservations and enable accurate estimations on future parking states. An efficient LAVP service framework enhanced with parking reservations is then proposed. Benefited from the intelligent predictions, parking load can be accurately predicted and greatly alleviated at individual car parks, thereby avoiding overcrowding effectively. Results show that significant performance gains can be achieved under the proposed scheme by comparing to other benchmarks, with respect to greatly reduced waiting duration for available parking space, as well as enhanced customer experiences in terms of reduced traveling period, etc. In particular, the number of parked vehicles across the network can be effectively balanced.