Have you done any benchmarking on the sizes of the images? I cannot imagine a JPEG math expression, cached by google, is going to take an age to load when viewed in Gmail and not be that large.
That's a good question. The actual size of the images is not large: the problem is that I can visually see the email take a while to load properly, even for an email containing just two PNG images that are 2941 bytes and 6333 bytes respectively. If nearly every sentence were to be interrupted a few times with such delayed-loading images, the reading experience would be rather poor.
(This was with the images self-contained in the email, which seems to be implemented under the hood by having them be attachments and referring to them in the body, so the loading happens out of order... I haven't tried with the images being loaded as external images, but that has its own problems such as Gmail probably not loading them by default, or having to ensure the images will be available over the internet for as long as anyone might want to read the email which if you're sending it to a mailing list is ideally forever, etc.)
Inline base64 shouldn't have any head of line blocking or slow rendering issues, unlike embedded images-as-mime-attachments, which will normally result in one HTTP request to gmails servers per image, and without QUIC you'll get head of line blocking and limited/no HTTP pipelining making that very slow.
Inline base64 should perform fine for small images up to ~10k images I'd guess on most hardware.
If your email contains more than 10k math equations, perhaps email isn't the right tool...
