I have a different hypothesis. Every feature you bolt on decreases the learnability of your language, decreases the navigability of your documentation (the books are now 20% thicker, and the old books and web pages are now useless), and decreases the odds that two programmers who "know Java" can actually read each other's code.
It is also very hard indeed to introduce a new feature while preserving the functionality of existing code. The gyrations and compromises that are required are often enough to destroy most of the value of the new feature. Better to build a new system around the feature and let it shine, not struggle to fold a bastardized version of it into the old system.
The time to tinker with your system is when it is young. When the system is old, and people have already built their lives around it, it is too late to remodel the foundation. You can take this principle too far [1] -- it is important to enjoy your youth fully, not settle into premature senescence -- but any system which becomes popular will sooner or later reach maturity.
Eventually your codebase, installed base, community, and documentation reach the point where every time you change something important, you lose more in coherency than you gain in power. [2] At that point your system is mature. The best thing you can do at that point is to accept the facts of life and let the system settle into its role -- preserve the value of the work you've already done, the books and lectures you've already written, the people you've already trained; then build on that work by calling its APIs -- while working on a new system on the side. Because it is, in fact, often better to have two or three smaller, comprehensible systems with fundamentally different features and maturity levels -- Erlang for the telephone exchange, Lisp for the airline-schedule AI, PHP for the three-line CRUD web page -- than a gigantic system that is constantly remodeling to add more corner cases, like the Winchester Mystery House.
Of course, Java's designers have never believed this hypothesis, or they would have left well enough alone years ago.
---
[1] As in the legend of the "make" syntax, which supposedly remains broken because, by the time the designers realized it was broken, they already had ten users and felt it was too late to fix. Don't worry about the first ten users. By the time you have a million users, however, it might be time to start thinking about stability.
[2] Think: Windows. It's mature. It no longer seems to be possible to make major changes in the fundamental structure of that operating system. And that's not necessarily a bad thing, unless Microsoft insists on trying, spending billions of dollars and most of a decade and breaking a lot of existing infrastructure in the process.
