R and multicore architectures?

We are gradually moving most of climate data and model output  
analyses to R. We have a 2 x Core 2 Duo Mac with 9GB memory and 3TB  
disk space, i.e. ideal for such work...

Question: Does the standard (binary distribution) of R use all of  
the cores and access all of the available memory?

The CRAN binary is a 32-bit build, so it will use only up to 4GB of  
RAM per process. An unofficial 64-bit binary of R is available from  
the R for Mac developer page, but CRAN doesn't provide binary  
packages for that build.
However, the CRAN build supports vecLib, and as such it will use all  
four cores for matrix operations that are supported by LAPACK.

Or does one have to recompile R to take advantages of more powerful  
hardware available around?

You don't have to, but you can compile 64-bit version of R if you  
need more RAM in one R process (you can still use all 9GB of RAM when  
using parallel R processes, c.f. snow). In Tiger the 64-bit version  
of R is somewhat limited compared to the 32-bit version, because most  
system libraries are not available in 64-bit, however Leopard  
supports full 64-bit build including the GUI. We will release a  
corresponding binary in due course.

Cheers,
Simon