The Internet of Things (IoT) is characterized by a wide penetration in the regular user's life through an increasing number of mobile Things, such as mobile phones hosting sensors and actuators. However, the shift to the mobile IoT does not come without challenges, as many already existing issues remain unresolved and are amplified by the IoT scale and the mobility of its Things. The most challenging issues are handling the abundance of users and Things, providing interoperability across the heterogeneous Things, and overcoming the unknown dynamic environment due to the mobility of Things. This paper addresses the above challenges as we revisit the commonly used Service-Oriented Architecture (SOA). This leads to the design, implementation and evaluation of MobIoT, a new service-oriented middleware. MobIoT modifies standard SOA functionalities, namely service discovery, composition and access, to better address the challenges posed by the IoT, especially its scale. Specifically, MobIoT adopts probabilistic methods to decrease the number of involved devices, while building on semantic knowledge to support interoperability and fulfill users' queries for Thing-based measurements/actions.