My guess is they were chunking the multiplier cascade and handling using parallel circuits so that the waveform could be independent across the chunks and not be dependent?

Edit: found it.

2240004 3-24 C (you will want to search by this key, it's a long document!)

http://ed-thelen.org/comp-hist/CRAY-1-HardRefMan/CRAY-1-HRM....

"Note that reversing the multiplier and multiplicand operands could cause slightly different results, that is, A x B is not necessarily the same as [B] x A"