The adoption of innovations, products and online services is commonly interpreted as a spreading process, which to a large extent is driven by social influence and mediated by the needs and capacities of individuals. Its modelling usually involves behavioural threshold mechanisms and predicts the evolution of explosive, system-wide adoption (known as a global cascade) if the system satisfies a set of theoretical conditions. However, these models fail to describe spreading in several real systems where, in spite of global adoption, the spreading does not follow a rapid cascading pattern, although threshold mechanisms play a role in their dynamics. In this work we propose a solution to this paradox through the analysis and modelling of online adoption in the world's largest voice over internet service, the social network of Skype. We provide empirical evidence about the distribution of behavioural thresholds and other structural and dynamical properties of this worldwide adoption process. Incorporating the observed features into a dynamical threshold model we show that the spread of online services is constrained by the fraction of non-adopters in the network. Such users who are not susceptible to product adoption can be considered as blocked nodes hindering the system into a quenched state where the evolution and structure of the global adoption cluster is very similar to the empirical observations. Model calculations and the analysis of the real social contagion process suggest that the structure of the adoption clusters differs radically from the case of rapid global cascading, since first-order triggering effects appear to be important only locally, while second-order and external effects are responsible for the emergence of global social adoption.