The cognitive load placed on users by both the proactive and spontaneous provisioning of service functionality and by the physical activities performed in ambient intelligence environments can lead to the depletion of their mental resources. This paper demonstrates how burdening the inappropriate selection of service functionality can be for users by conducting a semi-naturalistic and controlled user test to investigate the significance of the cognitive resource depletion problem in specific ambient intelligence environments. A dynamic service binding and scheduling mechanism is provided based on different types of interference and on mental resources and their demand requirements. A technical evaluation is conducted by simulating the mechanism over a set of various abstract service compositions, making use of real datasets of user interactions with diverse HCI services and daily physical activities. The results show that this mechanism ensures less cognitively taxing, unobtrusive service composition provisioning.