A model of prospective time estimation was tested in two experimental variations which examine the influence of load switch in task demands on time estimation. The model predicts these influences on time estimates by means of memory processes such as spreading activation. The approach was integrated into a cognitive architecture and has previously been tested successfully. In two experiments participants had to work on a counting task with different levels of working memory demands (High/Low). The participants had to stop each trial after a perceived duration of a previously presented sample of 100 seconds (altered reproduction method) and received feedback. In the Low group most trials were performed in low load and one or two trials in high load (load switch), and vice versa for the High group. For the Low group the model predicts overestimations at load switches, but underestimations for the High group. We found that the model predictions in the first experiment only match the experimental results for the Low group, most probably due to the experimental design. In the second experiment, the design was therefore slightly changed and the timing task was embedded into a manual control task within a microworld environment. In this setting the model predictions match the time estimates for both groups. The series of experiments reported give strong evidence that the model is able to capture and to predict influences of task demands on time-estimates. The timing model may be used as a base for modeling subjective temporal reasoning and the timing of interaction with a dynamic system.