Taking advantage of the latest features of any processor, including MMX technologies on Intel's (or anyone's) processors, is generally a matter of recompiling the relevant code with a compiler that knows about, and can optimize for, such technologies, of using libraries and routines which are optimized for those technologies, or of hand-writing code that uses those technologies.
You can be sure we will take advantage of all compiler optimizations and optimized libraries available to us when we release any version of the BeOS. Future releases of the BeOS will take advantage of even further optimizations, as they become available. And of course, where appropriate, we also hand-code in assembler to squeeze the last bit of juice out of a given processor.
For MMX specifically, we support all MMX instructions on MMX processors (that is, developers can write programs that use those instructions and they work fine). The BeOS itself does not (yet) explicitly use any of those instructions, but that just means it will get even faster in the future. We fully intend, for example, for Release 4 to be faster than Release 3.