A year or two ago I ran across an interesting study on exactly this that was done several years ago. I can't find the study right now (I'd appreciate it if anyone who knows the one I'm talking about could post it), but the idea was that they took a reasonably complex but well defined problem and had it implemented by a variety of programmers in a variety of languages (C and Perl were in the list). They then compared how long the solutions took to implement, and how rapidly they ran.

As you might expect, the most rapidly implemented solutions were the ones done in the scripting languages. The fastest solutions were in C.

As you might not expect, the slowest solutions were in C. Furthermore the variation both in time to implement and runtime performance were far larger in C than in scripting languages.

What seemed to happen is that the scripting languages provided an obvious good approach to the problem, and everyone took that with decent results. In C there were many choices to make, with more difficult to see consequences for time to implement and runtime performance. (Which is exactly what you are talking about having happen.)

I'd wager that Java would have resembled C in this particular test, but without the performance.