Kay claims that PARC hardware ran Smalltalk faster because the hardware was optimized for it. Sort of -- the Alto and successors like the Dolphin and Dorado had programmable microcode, and they implemented the Smalltalk VM at that level. But this is hardly a panacea, and the fact is that those machines had pretty poor performance (except for the Dorado, which solved the problem by throwing exotic ECL chips at it, at a cost of nearly $100,000 a unit.)
I used Smalltalk on most of the Xerox machines in the mid-'80s. Performance was acceptable but not good. (By comparison, I was stunned at how fast the original Mac was when I first saw it.) The first machine I ever used that ran Smalltalk at high speed was ... a Sun-3 workstation with a Smalltalk that translated bytecode to 68000 machine code. (This was one of the first just-in-time translators, written by Peter Deutsch.)
My point is that Kay is incorrect to claim that hardware was an issue in the choice of Smalltalk or C. I think a bigger reason few other people used Smalltalk was that the environment insisted on taking over the entire computer, and would not co-exist with any other software. C, by comparison, would fit in anywhere.