When reading from stdin, mqlrun makes small, unbuffered reads that hurt performance with small documents. A spike solution using buffered reads increases small-document performance 2x to 5x depending on the input volume.