When is it sensible not to use XML?

This paper considers the problem of comparing the read time for a file containing an array of one-half million one-byte data values using a direct read time with the time required to read an equivalent XML file. The XML file uses a variable-length string encoding of the identifiers associated with each of the byte values. The XML file is about 180 times larger than the original file of byte values. Because the XML file is much larger and because the XML parsing requires substantially more computational work, reading the XML file on a Windows-based PC takes about 3,700 times longer than the direct read. The experimental study shows that the XML read time increases linearly with the size of the file. Given the increased read time and the increased probability of programming errors for the XML encoding, it appears that in cases where data archives are providing simple binary or text files of data that may be readily understood and read with short programs, then XML may not be an optimal choice for serving users or for preserving data.