Due to the good off-line results (in terms of high learning rate, high success rate, low false alarm rate and high prediction probability) provided by Venn machines as adaptive disruption predictors in JET, this article assesses their use under real-time requirements. Venn machines predictions can be expensive from a computational point of view but they can be used in JET in a deterministic way. The JET characteristic time to take mitigation actions for disruptions is about 30ms. This article shows that Venn predictions take computation times of 1ms for JET conditions. The influence of both the dimensionality of feature vectors and the number of training examples to make predictions are analyzed. Also a short discussion about the potential applicability to ITER is presented.