This paper aims to experimentally evaluate the performance of one popular algorithm for the detection of tampered JPEG images: the algorithm by Lin et al. [1]. We developed a reference implementation for this algorithm and performed a deep experimental analysis, by measuring its performance when applied to the images of the CASIA TIDE public dataset, the de facto standard for the experimental analysis of this family of algorithms. Our first results were very positive, thus confirming the good performance of this algorithm. However, a closer inspection revealed the existence of an unexpected anomaly in a consistent part of the images of the CASIA TIDE dataset that may have influenced our results as well as the results of previous studies conducted using this dataset. By taking advantage of this anomaly, we were able to develop a variant of the original algorithm which exhibited better performance on the same dataset.