To be clear, I'm not generally a fan of infinite scrolling either.
I'm not sure it is possible to talk about this in general terms - the specific use-case needs to be considered. In the case of Wikipedia, I think that the page size of a typical Wikipedia page isn't problematic.
I suspect it's actually pretty rare that the HTTP response size for text documents is big enough to be a problem. 1M of text is pretty large in terms of text, and most pages would have more than that in embedded images.
1. The desire to limit the size of HTTP responses. 2. The lack of the ability to notify a browser that you're done loading/rendering async content.
Aside from page-flip-metaphor apps like the one posted here, it doesn't seem like a desirable UX pattern.
Have there been recent (or not so recent) developments in either of those problem areas?