I found this paragraph particularly interesting, I hadn't thought about this aspect before
It's interesting that pianists whose playing had less rubato (Rachmaninoff and Lhevinne for example) tend to fare better on rolls than those who played with more rhythmic freedom (Paderewski and Friedman, whose piano rolls are ghastly). This has to do with the fact that rubato and sound are inextricably linked. You can't take the timing of a rubato and separate it from the nuance of a rubato and have anything other than a mess. When working at the turn of a phrase in, say, a Chopin mazurka we are literally splitting hairs of inflection and colour. If that F sharp is played a quarter-second later it will need a slightly different weight of sound. To hear it inadequately on a '78' is frustrating, but true; to hear it approximated on a piano roll (on a different piano, different hammers, different strings, different dampers, different soundboard, different rim, different keybed, different repetition action etc.) is a travesty based on a total fiction.