The problem is twofold: 1.) Someone who thinks we need more programming languages doesn't understand the problem, but it's generally a moot point because 2.) Someone implementing the latest fad in languages inevitably ends up re-inventing the wheel, albeit poorly.
The end result is that while 10% of the features of the new language may be novel and interesting, only about 1% of the features will be useful in the practical sense. And 90% of the time spent in language development will be put into recreating existing functionality in the new language. Look at Java, for example. Just when UI design was starting to make sense, Java reimplemented the entire GUI in Java, and threw away a large portion of what had been learned on how to do UIs well.
And the result? Nobody used the Java UI. Remember Swing? Yeah, I actually wrote GUI apps with it, but Java went the route of the server - hidden from the user, and never ever came close to challenging C++ for work on Desktop apps. Oh, sure, maybe Swing (or whatever they're calling it these days) has improved, but Java missed an entire generation of programmers in the interim.
The problem today is that by the time your new paradigm catches on, and you've developed all of the attending libraries and functionality to make it competitive with existing languages, the paradigm is already obsolete.
There really has been very little progress on languages since C++. There's very little that's new, functionally speaking. The new languages of today aren't revolutionizing the software paradigm - more often than not, they're just reimplementations of existing functionality and language paradigms with a new set of strange and irritating behavior. Python's use of whitespace to perform block delimiting is an interesting step backwards - it was first done in COBOL - and it was just as much a defect then as it is today. When I had to learn COBOL in school, I thought it a waste of time until Python came out, and then I recognized the value: it had taught me the danger of making stupid decisions in language design.
We've come to the point where unless you're using a really old language - older than C - the language is pretty much irrelevant. Granted, some languages are better than others at solving certain problems, but I've yet to find a new language (after Java, that is) that brought something new and important to the realm of computer science. Even Java was a draw - you lost operator overloading and multiple inheritance - in favor of a simpler, more practical paradigm (sound like C, anyone?!). And C# was definitely a step backward; you got none of the advantages of a vm with all of the disadvantages. And Python has just decided to break backward compatibility on a whim, leaving anyone with a substantial codebase out in the cold.