I was using lame binary from fink distribution. It gives from 0.4x to 0.6x performance in G4/866 Dual(without --multi. --alt-preset standard). So slow, I tried to recompile lame 3.93.1 with newest apple gcc-3.3 (can be obtained through ADC) and set CFLAGS option to "-O3 -faltivec -mcpu=G4" to take advantage of altivec optimization. The result is 3times faster than the fink binary. Recompiled lame gives 1.0x to 1.5x perfomance (with no --multi) Try it !