I agree, and the memory timings in both his settings are the same which is comparably fast, which don't give an upper hand to the lower clocked system , also if the memory is operating in dual channel mode then the system as a whole will hardly notice or react to an extra 200 to 500 MB/s of mem bandwidth or whatever he's getting (probably 200-300 which can't compare to the total system memory bandwidth and can be concidered barely 250MB/s more bandwidth if you take into account the fact that the 196mhz mem system will likely benefit greater from the increased fsb of 245 than the already plentiful default dual channel settings of 200 1:1 will from a modest pump in the fsb and mem to 220 ).
basically what i'm trying to say in the parenthesis is that the cpu when running over 3ghz with 5:4 will have an increased need for bandwidth to and from the cpu because it already has a ratio of roughly 5:8 fsb:mem bandwidth (dual channel mmm) which means in most cases it'll fall well above a single channel solution whose demand for increase memory mhz might matter more. The fact that dual channel versus single channel never really yeilds more than 10% in real world situations leads me to believe that the slight hit at 5:4 will not harm the system more than maybe 2 to 3% at the same clockspeed imo (atleast under cpu-only tests) and the fsb increases and clock speed increase each by themselves are more than enough to account for the slight discrepency and then some with change to spare
my guess would be that the two systems might come closest (not close enough tho) in performance scores when put in situations that rely heavily on the vid card/agp bus because this is the only situation i could see the extra bandwidth really helping and possible saturation of the fsb and mem bandwidth at the same time with prolonged maximum intensity might give the 1:1 system with more mem bandwidth the upper edge when it comes to those extra 1 to 5 fps that the video card can push out. Also it is possible that in a well programmed and highly "new" benchmark might find the system to be cpu-limited because either the vid card is that good or the game/benchmark uses intense AI/physics/calculations which bring any cpu between would need to be complex enough to bring 3 to 4ghz P4's or slower to a crawl just on intense work.
This situation is highly unlikely, and as almost all games/graphics and full system gaming benchmarks i've seen seem to not pressure the cpu nearly that much before maxing out the vid card, whatever kind it may be, because using the resources in this way is neither neccesarry or likely to be a goal of any software/game/bench/etc,.
Therefore I would stake 5 bucks on the 1:1 system not winning any benchmarks unless they are synthetic mem bandwidth benchmarks, or some kinda program that needs access to the memory from places other than the cpu, and it's fsb, and with intense consistency/bandwidth too (the only real candidate for this is the vid card and the 8x agp bus tho).
the cpu is still the main peice of the computer when utilized in any program (including games and gaming benchmarks), as much as nvidia and ati might like to see different in the future of desktop graphics and "home pc media/graphics centers" or whatever they might call a pc whose most powerfull and often utilized for "normal daily use" single computing unit is the video card (caugh did someone say console).
Because of all this i believe the fsb increase is nearly twice as vital in this particular situation as the memory increase in almost all situations/uses unless other factors are involved. The "substantially higher clock speed" of the cpu, when compared to the the nearly two or so P4 models higher in the performance list that this represents, the answer becomes clear
go 3.6 go !!!