Which, according to reports, back when, only gains one a few percent performance. Also, it only applies when using a G3 processor. Still it's some improvement. However, there was also a report that PCI performance was lower on the Kansas boards. At this late date I'm not sure I'd trust either report without some carefully controlled test comparisons.
One thing to remember is that Tsunami and Kansas are not *different* architectures. They are very slightly different implementations of the same "Power Surge" architecture. They use the same Hammerhead bus arbiter/memory controller chip on the CPU bus, the same pair of memory data controller chips ( slightly different on the ANS with its support for parity memory), the same PCI bridge chips (bandit), the same interrupt handler (grand central), and all the same peripheral chips which provide SCSI, Ethernet, serial port, etc..
The only hardware differences Kansas boards have is a lack of on-board CPU cache (9600) or cache socket (8600), because the cache was moved to the Mach V CPU card; different ROM version, revision numbers I don't remember but they're in my referenced post, but part numbers on the four ROM chips are 341S0280 through 341S0283 (original) vs. 341S0380 through 341S0383 on the Kansas boards; and some slight wiring differences in the CPU socket which I think are just an expansion of the number of 3.3V pins, but I never traced it out in enough detail to be willing to say with certainty.
Replacing some of the 12V or 5V pins with 3.3V would be plenty to make the CPU card's not compatible, yet be fairly easy for 3rd party upgrade makers to adapt around.
If one takes a set of Kansas ROM chips off of a Kansas board (or programs the contents into 44 pin PSOP Flash chips, e.g.) the Kansas ROM can be installed in all earlier x500 and x600 machines and clones and they work fine. Strangely it also works just fine on the 7200 and Catalyst clones but not that strange given that the Catalyst systems use the same ROM as the early x500 machines.