How can one test the quality of ICC profiles? I have been looking at the profiles of laser and inkjet printers and find the vendor-supplied profiles vary in size from 37KB to 2MB, from 8 nodes per side of the CLUT to 33 nodes per side. Some use 8-bit values, some 16.

Are there tests or evaluations that allow one to determine the quality of these profiles other than just eyeballing the results? Are there guidelines on which of these variables are most important to color management?

All other things being equal, a fully 16 bit profile is preferable, meaning that a large printer profile is better than a small printer profile. For instance, the greater precision makes for lesser impact from white point problems in ColorSync 2.6.

The Kodak ICM1 implementation only supported 8 grid points while the Heidelberg ICM2 implementation gives the same accuracy as on Mac OS.

You don't want to use the older ICM1 framework nor profiles made for it.

Still on the quantitative side, if you are working with Photoshop the BtoA0 Perceptual and BtoA1 Relative Colorimetric tables should have the same total area coverage.

The black generation controls are different from one print profiler to another. The print profilers in actual use are Printopen and ProfileMaker.

As for the 'color kerning' of the Perceptual tables, then each print profiler has its own approach. There is no universal answer to what is good and what is bad.