Mozilla.
Graphics is loaded once, true. But HTML is unique for each post.

In this thread for 500 bytes of your post there's 35K of HTML with all the other data. So for each 2K or so post I need to load 35K or so of data repeated elsewhere, but since the HTML is unique it isn't useful cached.

True, 35K is not much and the HTML of 8052.com is quite nicely written, but if a thread has 20 posts, 2K long on average, that would make 35K+40K=75K download in one chunk, and not (35K+2K)x20=740K in 20 clicks to read the whole thread.
Of course not everyone reads the whole thread, but you need 2 new posts to nearly break equal (2x37K) versus reading whole thread (1x75K).

I guess a sensible technique to limit a page length would make sense - i.e. 30 posts per page, or 100K page length limit. But a single post in a web-based forum where you need to wait for each new post to load is too small unit of information for comfortable reading.