The key thing is to develop new techniques for solving problems - think iterative versus recursive, or functional programming versus logic programming.
A given language is most likely targeted (or has least resistance) towards one style of programming, which means that new abstractions are likely to work better in a new language, one that has been developed with these blocks as starting points rather than added later.
It might be nice to focus development of specific key languages to avoid too much fragmentation by picking best-of-breed, but everyone's got their own itch... and for some that is developing new languages.
That's why I welcome the creation of any and all new languages. I don't need to be on the cutting edge and try to learn all of them, I can wait until the early adopters see value in them and then learn from the things that work.
Do you want a company to 1) only brew products (with a business model) internally until it is ready to ship, or 2) develop and release lots of things and see what works or not?
Google leans towards #2. They don't really care if they can make money off of something, if one of their engineers is interested in something they're allowed to take that risk. They put lots of stuff in the wild and if it doesn't stick then they kill it (or in most cases leave the source to the public - see wave) - or try to make it profitable (look at gapps).
Yes, a lot of things coming out of Google die, but that's only because they're putting out lots of things all the time. The day they stop putting out unproven stuff and only release v1.0 products is the day they lose their innovation edge.