I mostly use those kinds of online sources, because I'm usually on the internet and rarely in libraries, just in terms of proportion of my time, and I rather strongly disagree that those sources are anywhere near as good as a library, at least for what I'm looking for. It's a huge breath of fresh air to find a good book on a subject, because it more often than not actually covers what's going on in a way that is far more opaque in dozens of StackExchange posts, blog posts, mailing list archives, etc. The rational exposition is just much better, if you can find a good book: explaining not only what's going on but also why and how, how it connects to other things, how it came about, where to look for more information, etc. It's really frustrating to work without access to a good research library for that reason, because you get only these much more scattered bits and pieces on many topics.
Two areas that I've come across semi-recently where this is particularly true: DSP and logic programming. There are some good books on those subjects, and they are very much better than trying to piece together Google results, especially trying to learn an area.
I guess a lot of it depends on how you learn best. I like to go deep, but it has to be on my terms. Sucking in books never gives me a sense of accomplishment unless I'm actively trying to 'rediscover' what is being taught for myself. I prefer to use reference material to open up lines of thought, then I go away and try to 'reinvent' it for myself, baby steps. I come back later to have the material nudge me in the right direction or get me over hurdles. Sometimes it's too hard and I'll drop material for something else, even if it's not as deep, or go without, rather than grind on, because I'd rather not know than 'know' and not truly understand.
Fully-formed 'rational exposition' of a new topic or area of study is just a terrifying thing for me because I'm worried I'll be swept up in all that exposition and not prod the holes in my understanding, perhaps not even see them. Later on I then feel I have to remember 'what the author taught me' rather than unravelling of the topic itself. Visiting a complete exposition later, after a bit of folly, I feel much more confident in both myself and the material. It's very inefficient in terms of time, but then I can spend a lifetime learning and still know next to nothing anyway.
Two areas that I've come across semi-recently where this is particularly true: DSP and logic programming. There are some good books on those subjects, and they are very much better than trying to piece together Google results, especially trying to learn an area.