Stream clustering is a technique capable of identifying homogeneous groups of observations that continuously arrive in a digital stream. In this work, we inherently refine a TF-IDF-based text stream clustering algorithm by the introduction of an automated distance threshold adaption technique for document insertion and cluster merging, improving the performance during distributional changes in the data stream. By conducting a thorough evaluation study, we show that our new fast approach outperforms state-of-the-art one-pass and batch-based stream clustering algorithms on various existing benchmarking datasets as well as a newly introduced dataset that poses additional challenges to the community. Moreover, we find that current evaluation approaches in the field of textual stream clustering are not adequate for a sound clustering performance assessment of evolving distributions. We thus demand the utilization of time-based evaluation.