I got the impression from the spec that it was originally designed to support retaining the compression dictionary across frames, which would have produced much

I got the impression from the spec that it was originally designed to support retaining the compression dictionary across frames, which would have produced much smaller files at no cost to the decoder. But this wasn't explicit, it didn't get implemented, and the second spec ruled it out. It could have very easily been better. I'm not sure what the moral is.