[JFreak]

Well this isn't a technical explanation; but imagine PT rendering your output one minute before realtime (when you start to hear click in the cans), which would mean it has already calculated whatever you need to hear in the cans while recording. That would mean the software only needs to stream one feed from memory and record what you want to record, so therefore recording buffer can be lowest possible.

Now replace that "one minute" with a reasonable amount of time (say, a second?) between hitting record and hearing the click and we're almost there.

Of course, this is not how it works because you can adjust levels realtime, but to understand the concept of dual buffers this is as close as it gets if we talk layman's terms.