As far as I know, a language with less keywords lends itself to a faster parser. That's the motivation behind using "auto" for type inference in C++0x, "auto" is already a keyword.

The difference is negligible. Complexity of grammar (not the same thing) is probably the big cost factor. The big thing there is usually ambiguity of a statement within a single context that has to be resolved. Static is not such a case because there are three distinct contexts. But when a parser has to read many tokens ahead just to figure out where you're going with a term, that's what slows things down.