I'm not quite sure I agree that if Facebook lost 1 out of 1,000 posts that no one would care. I do agree that with Google most people wouldn't care if they got slightly different search results with the same terms. It's the difference between instantaneous and persisted data. Facebook posts are essentially persisted data. They get put in and are there for ever. Google searches are more transitory. With Google continually crawling the web if a page gets missed once it'll get hit and put in again later so it's not a big deal. Neither of these are critical but it's the nature of how the data is generated and used that makes the difference.
I'm not sure how many users monitor their guest book or blog posts close enough on a daily basis to notice if one (out of a couple hundred) entries from months back suddenly disappeared. I'm sure somebody eventually would, and they'd be really verklempt about it. However, there generally isn't something like a Service Level Agreement between a social media company and their users. Even if the issue were brought to the company's attention, I doubt they would respond by assigning a DBA with the task of digging though backups or transaction logs to locate the missing data.
On the other hand, if a bank were dropping transactions, within a few hours customers would start calling in with complaints about non-posted paychecks or missing daily deposits. It would become news really fast, and the bank would be required by law to fix it.
Regarding where NoSQL databases can properly fit in a corporate enterprise envrionment, there is a lot of non-transactional stuff like documents, images, reference data, and entity-attribute-value records that could be better offloaded from the RDMS into NoSQL. I could see the merits of a blended architecture, even in an organization like a bank.
"The universe is complicated and for the most part beyond your control, but your life is only as complicated as you choose it to be."