Jim Hugunin has posted a paper on IronPython, a fast Python implementation for .NET. This has triggered a cascade of posts.
The money quote: "[A]s I carried out my experiments I found the CLR to be a surprisingly good target for dynamic languages...."
More: "High system performance is the end result of hundreds of small decisions rather than a single large one. Much of IronPython's performance comes from careful consideration of performance in each design choice...."
Essentially, instead of "just" getting the semantics of the language right, Hugunin strove to use native CLR facilities whenever appropriate and provided alternate "fast-paths" for common situations (such as for function calls with a fixed number of arguments) while providing for the more general solution with slower code.
While compiler writing is the rocket-science of computer programming (although game programming comes close...) Hugunin's tactics don't seem unreasonably burdensome.