If you're building a platform, yes, you should at least be testing your point releases against a handful of the most popular applications that run on your platform, because things like this, even if caused by bad decisions in those 3rd party applications, look bad against the platform as a whole.
If this was a big feature release like JDK7, that's one thing, but a bugfix release is completely different.
Edit: Everybody would be grabbing for their pitchforks if Facebook all of a sudden updated their backend and broke FarmVille. It's the same concept.
Just to add onto this: a company I worked for was a somewhat heavy user of a 3rd party api, and they integrated our key use cases into their test scenarios in order to avoid breakage.
Depending upon what percentage of Java developers use eclipse, seeing if it still works seems fairly reasonable.
No, Oracle should be responsible for not breaking things with no warning that a large fraction of the customers use--it is not relevant whether or not those are third party things. A huge number of Java developers use Eclipse, and so it should be something Oracle tests with--that's just basic QA that any competent company should do.
Then, when they find there is a problem, they should notify the Eclipse people. If Eclipse isn't fixed in time for Oracle's release deadline, they should at the very least put something in the release notes that there is a problem.
Well - when you're at some serious size, it makes sense... Kind of like special-casing the memory allocator in windows if SimCity is running: http://www.joelonsoftware.com/articles/APIWar.html - how could they not be compatible with SimCity?
And Oracle really should have tested stuff, even if Eclipse did do something stupid.