What's your point with this article? I suppose you are advocating that Sun desists of maintaining backwards compatibility, preferring to remove old APIs (or add enhancements that require breaking changes)? If this is your idea, the bad news is that the gain would be exceedingly small. Summing all of the bad, old and ugly APIs that would be good candidates for removal would deliver a very small fraction of the latest JRE's size, and also of other costs like continued development and porting. The only exception would be enhancements that need a HUGE compatibility break, e.g. an alternative generics implementation without any compromises to keep old apps (that need non-generified collections etc) binary-compatible with new JREs. But upgrades like this would amount to a different language and massive porting or rewriting efforts for existing code; if you can cope with that, there are options like Groovy, Scala, JRuby etc.

Reality is that the Java compatibility story is exceptional. I have a couple large Java apps that I wrote by 1997-98, didn't receive any maintenance since ~2000 as the project was discontinued, and every year or two I perform a quick preventive maintenance - check out the code from repository, build it with the latest and greatest JDK and IDE, run a quick test - and everything continues working... the number of compiler warnings (like deprecation) increases slowly, the code looks more horrible compared to modern practices (e.g. hundreds of JSP 1.0 files stuffed with scriptlets and no MVC framework at all)... but it works.

And of course I'm not the only programmer that can code properly, there are zillions of Java apps and libraries out there that run reliably on a large number of JRE versions and OS/HW platforms. Yes, this is more common for free & open source apps, simply because FOSS developers usually care about user choice. If you complain that some app doesn't run on a different JRE, a typical commercial vendor's response may be "it's not certified on that JRE, please use the one we ship with the app". But an open source group will answer by fixing the bug (and yes, 90% of the time it IS an application bug) or provide a workaround. BTW, such bugs are typically caught early in FOSS projects, due to the habit of providing source distributions and letting users do their own compilation - with whatever OS, runtimes and compilers they have installed. So the developers get multiplatform testing for free (which is good, because too many FOSS hackers only use things like Linux and will rather touch fresh cat shit than a proprietary OS).