Calculating term frequencies in a big corpus efficiently regardless of document boundaries

Question

I have a corpus of almost 2m documents. I want to calculate the term frequencies of the terms in the whole corpus, regardless of document boundaries.

A naive approach