I think the quote implies that Feynman's suggested changes were for marketing materials. It turns out myself and my colleagues were the intended targets of those marketing materials. We ignored them and just benchmarked the system.

In the end none of it mattered because it was next to impossible to achieve significant fractions of the quoted "peak" performance for real world/non-embarrassingly parallel algorithms. Especially with the CM-2 and 64 bit floating point.