99,999% of MTA bounces can be automatically classified, there's not so many different MTA implementations in the wild. The same applies to 99,9% of vacation responders because most people use the standard templates from their Outlook or Webmailer.

The rest is intern-food. A single person can easily classify 20+ mails per minute. And that's what happens in companies with real customer service.

Uh, have you ever actually tried? Because I have, and neither of those numbers are anywhere close to what I experienced. It's probably true that a plurality of the messages are standard Outlook autoresponders, but not "most" -- and definitely not 99.9%. And a very large number of MTAs are either misconfigured or intentional subvert the standard in order to confuse spammers.

These days there are thousands of "So and So doesn't work here any more" messages that have to be removed from the list by hand, then the people using Lotus Notes, Eudora, and god knows what else. Also non-standard Outlook messages and Outlook messages in different languages. Worst is those whitelist anti-spam challenges that require you to fill out a CAPTCHA (or pay the spam filtering company a bribe!) for each message you want to send to a subscriber.

We do comb through the replies because email is an important aspect of our business, but it's a lot more work than you make it out to be. If you're just a small company sending out a monthly newsletter, I'd probably say don't bother.