Well, to me that's a good argument for continuing to collapse (or otherwise avoid rendering) the lower sections -- they're less likely to be read at all, so why waste CPU time and bandwidth on them? My typical Wikipedia inquiries are resolved primarily by the first paragraph or infobox, and maybe skimming one or two sections whose titles sounded relevant...

Of course since we're transferring the text data anyway, it may not save much resources to just collapse/hide the text as we do now. Smarter behavior on lazy-loading of images might accomplish the bandwidth goals even better...

Another thing to consider is that if we don't collapse the intermediate sections, and we don't also devise an alternate table of contents, people may never get to the later sections because it takes a LOT of scrolling to get to them. Yes, you can collapse each section one by one and not have to scroll much, but that strikes me a more difficult interaction to discover than expanding each section in turn...