For my extremely minimal site, the RSS feed is still smaller than normal pageloads because it only has graphics by reference. RSS readers scrape surprisingly often, but the file is small and highly cacheable, and most of them do a HEAD request only initially and never ask for the file if it hasn't changed. Despite accounting for a large percentage of requests, RSS clients account for only a small portion of traffic. The cool thing is that a lot of these RSS readers are shared (commercial feed readers, ttrss instances, etc) and so they are caching the feed on their end and actually saving me bandwidth vs an equivalent number of users accessing the site directly - a lot of these put the subscriber account in their UA string so I can see that e.g. Feedly is making regular requests during the day but has 30-some users behind those requests. It's a better deal from a cost-perspective than those 30-some people checking each day.
So a) no, it doesn't contain images, just URLs to images where necessary; an RSS or ATOM feed is just a text XML document adhering to a certain schema, and b) the feed file is only as big as the site decides it needs to be.
For example, a typical blog or news organization could generate a feed containing the last 24 hours worth of new content. Feed readers will poll that and pull in new content while ignoring existing content.
In short: this should be a non-issue unless you're turning out massive amounts of new material on a regular basis, in which case you're probably a major news organization and can afford it.
> As a result, some blogs (like christine.website) stopped serving article contents in the RSS feeds.
This isn't because of bandwidth concerns. This is because they can't advertise in an RSS feed so they want to drive eyeballs to the site.
Personally, I run ttrss and use a plugin that scrapes the content from the source site and embeds it right in my feed. As a result I only have to leave my feed reader if I really want to for some reason (e.g. to read comments on the originating site... like, say, HN!)
[0]: https://github.com/WordPress/WordPress/blob/270f2011f8ec7265...
[1]: https://github.com/s9y/Serendipity/blob/e2044472c202a8368774...
The real reason why sites started switching to partial content is advertising. This is why personal or corporate blogs usually provide full content and some sites like ArsTechnica.com offer it to paid subscribers.
Is it a problem in practice (cannot be solved by removing images or similar)? Could it be improved through smarter polling or some "last updated since" header? Or by using an "active" RPC-like protocol where the reader specifically queries for "list of all articles since this timestamp" and can fetch them together or individually, with or without images?