In reference to the last post on this topic, there are a few things that back up Alan's comments in the thread. If dynamic languages can't scale, then an awful lot of people at Amazon, Google, and EBay are doing something very wrong. There's heavy use of "scripting" languages at all three of those, and - the last time I looked - their scaling needs are above and beyond the needs of almost every web project that falls into the so called "Enterprise" space.
In Gosling's follow up, he tries to wave away the "most time is spent in I/O argument", and well he should try to wave it off - because it invalidates most of the rest of what he says. The reality is, most web projects spend a lot of time grabbing results from the database, massaging them for display, grabbing results from the submission, and dropping them into the database. Being dramatically faster at floating point math doesn't mean anything for that kind of thing. For situations where it does, sure - pick the right tool for the job. The mistake Gosling and others make is of extension - since Java is faster for a few things like math, they think it should be used everywhere for consistency.
Heck, if that's his argument, he should just pick Fortran.