More cache is the useful most useful in multi-processor configs. Because you have more cache, the cpu spends less time accessing the system memory (with multi-procs which usually have some kind of shared system memory, teh result can be huge). For instance, Sun Ultra Sparc IIIe's with 8MB cache from 4-8 processor nearly retain a 90% efficiency while Pentium III Xeons with 2MB cache from 4-8 processor configurations are only about a 60% efficiency (based on Aceshardware's average and specINT/FPU performance).
From single processor performance, just look at Willamette Celeron (128KB), Willamettes (256KB), and Northwoods (512KB) at 1.8Ghz, since they're all available at 1.8Ghz.