With regards to speed, when you turn all the high quality options in lame you enable all sorts of paranoid acoustic analysis, probably much more than xing is doing. That slows it down lots.

Secondly, lame is designed for portability; it works on lots of different OSes, and lots of different CPUs. If you just want speed on the x86 platform, try GoGo. It a varient of lame targeted at top speed on the x86 platform. It has assembly optimisations, as well as MMX, 3DNow, SSE, and Athlon enhanced 3DNow optimisations. It's fast!