If you are sending a marketing email to a person, an easy image that has unique data is their name. :) And I didn't even think hard on this. Pretty much any content that could have been easily done as text is easily done as a "dynamic" image.
Now, they can detect duplicates, but only after they have gotten the contents. Unless I am mistaken on anything. (highly possible.)
Such instances can be detected (e.g. if you see 100 emails with the same HTML, but with image URLs that are slightly different). Google can then prefetch those images or it can re-enable the optin for displaying just for those emails.
If I were to do email tracking, I would just filter the GMail accounts out of the statistics, because you can't be sure of when GMail's proxy loads those images and what you're interested in is the conversion rate (not in the total number of people that opened their emails, you only care about totals for emails sent and clicks). But a service like MailChimp is not interested in doing this, because MailChimp is a third-party that's interested in showing big numbers to their customers.
And putting these numbers aside, the privacy issues related to IP tracking, or the security issues are gone. So I think this is good.
Certainly whac-a-mole. Didn't mean to imply otherwise.
For myself, no need to filter them out without evidence. Keep a few controlled accounts to periodically try and see what the delay is. And get extra suspicious if all images are opened at once on a mass send out.
Now, they can detect duplicates, but only after they have gotten the contents. Unless I am mistaken on anything. (highly possible.)