The thing is that the files being read are big enough that two different files very likely sit on different physical tracks on the hard drive. If you read one file, then the next file, then the next one, sequentially, the amount of drive head movement is minimized. If you read two, three, four, or ten files in parallel, the drive head has to shift back and forth a lot. Also, the buffering is less effective, since the drive reading ahead and filling the buffer is probably going to not fill the buffer with data that will be useful to the next request coming in from a different forked child. So the forks are actually working against each other, losing all buffering benefit, and even causing the drive to have to seek to and fro repeatedly.
When multiple processes are using the same physical resource to grab information that is distributed all over the place on that resource, in what amounts to be unpredictable order, it's no surprise that they degrade performance.
