Every time I get hold of a new machine, I run my cheesy, non-scientific Java benchmark on it and put the results in a file that I keep. I've got a decent selection of platforms now. Here's the latest:

Int     Double  String  IO      Total   Platform
-----   ------  ------  -----   ------  -----------------------------------------------------
0.632   1.092   2.048   0.655   4.427   Sun Ultra 10/440 (128M RAM, Java 1.2.1)
0.104   0.726   3.507   0.516   4.853   Apple Macintosh G4 (dual 877mhz, 512M, Java 1.4.1)
0.549   1.425   2.753   0.67    5.397   Sun Ultra 5/400 (128M RAM, Java 1.3)
0.16    2.31    3.07    0.33    5.87    P700 Desktop, 256M, Java 1.3
0.266   3.547   3.719   0.515   8.047   Gateway E5200 (Dual PIII 450, 128 Megs RAM, Java 1.3)
1.102   2.86    3.296   1.15    8.408   Ultra 1 200E, 256M, Java 1.2.1
0.157   1.14    6.557   0.767   8.621   Apple PowerBook 550 (256M RAM, Java 1.4.1)
0.261   0.982   7.687   0.588   9.518   Apple iBook 900 (640M RAM, Java 1.4.1)
0.07    0.06    10.065  2.423   12.618  P600 Celeron Laptop, 256M, Java 1.2.2
4.825   19.903  1.439   0.89    27.057  HP 9000/800/N4000 2x440, 4G (under heavy load)
2.207   5.423   22.302  4.305   34.24   Sparc 10 (2x50mhz), 128M, Java 1.2
0.895   18.134  62.011  3.782   84.822  Sharp Zaurus SL-5500, Jeode VM

(sorted by Total, lower numbers better)

Note that the Total can be swayed greatly by just one test, for example, the numeric benchmarks on the HP 9000 and the String performance on the Macs.