From your code so far, you read 1 million lines and then use a very inefficient method (lot of CPU and lot of memory) to make a string variable of those lines.

I don't understand exactly what you mean by 100% CPU? My Windows machine has 4 cores which essentially means 4 CPU's that share a common big memory space.

Unix is a time sharing O/S. Other processes will get CPU time even if one process is completely compute bound.

I am not sure about these various XML Perl libs, but every time that your program runs an I/O operation, the O/S scheduler will run. Maybe use something that takes less memory and does more I/O? Every time you do an I/O operation, the O/S scheduler will run.