Why is reading TextLineDataset with interleave still so slow?

Question

I have 5 text files where each line is just a sentence.

I am reading these file using tf.data.TextLineDataset and then I use a tf.py_function