I worked at a company developing proprietary image processing SDKs (more like libraries) for .NET/ActiveX. We actually had very extensive QA, probably >100% code coverage (at least 1 test for most code paths) in unit and functional tests, but it's really difficult to predict how customers are going to use your product.
For this reason, newer employees had to do a cycle of QA/support/dev, where you found out that there would always be one customer whose entire use case was the one case you didn't test for, and that makes your toolkit fail horribly. I'm not excusing these bugs, and Facebook is certainly a more complex platform to build on top of, but I can certainly see how a QA team could be working nonstop and appearing to accomplish nothing.
(Also note that since leaving this job better testing tools have come out/become more popular. The Ruby community especially is very oriented toward automating "mutate this test until it fails".)