Would like a bit more clarification, please Manfred. Like George, I'm a bit puzzled.
I guess it's because gamuts are normally expressed in a different color space than RGB, say CIELAB.
For example, in another forum I just converted one 8-bit step in one RGB channel and found that step to be less than the CIELAB 'Just Noticeable Difference' (JND), i.e. 0.7 vs 1.0 CIELAB unit. As you likely know, a JND is a Euclidian distance in a 3D gamut diagram. Some even say that one JND is 2.3 CIELAB units.
Perhaps the differences being discussed here are to do with large-ish round-trip edits in a non-parametric editor and are more noticeable.
*** time for a new computer, Manfred, my cheap 64-bit Dell is working quite happily in 32-bit floating point . . beware the trap, different meanings for "bit"