Back when the Techpowerup review of the 5090 FE came out, I found this picture of the fused-off SMs to be interesting. You can see that there's an entire GPC fused off, but also 6 SMs from 3 GPCs, leading to those 3 being weaker than the rest. I thought this might came out in some microbenchmarks, but it didn't so I put it out of my mind.

Until the recent debacle with some ROPs being disabled reminded me of it. You see, if you look at page 9 of the Nvidia Blackwell White Paper you'll see that each GPC has a 1:1 ratio between SMs and ROPs.
Nvidia sadly doesn't go into details over how the SMs are connected to the 2 ROP partitions, but it's not difficult to imagine how fusing off the 6 SMs could in some cases affect access to 6 ROPs by the rest of the SMs, depending on how said connection is implemented.

That's it, that's my bit of information and speculation, I hope people with more knowledge than me in GPU architecture can weigh in on whether this is a sensible theory or it's pure coincidence that 6 ROPs are disabled when 6 SMs are fused off.