well for that you would need 2 nf200 or plx chips actually, since every of those chips can only connect to 2 16x devices and make them share one 16x lane input link.
And looking at results from several sources it seems that nf200 is slowing down sli 1-3% instead of speeding it up... interestingly xfire seems to be slightly faster with nf200 hahah but that could as well be caused by different tweaking of the bios or the hardware of different boards, since there is no board that has a version with and a version without nf200 its hard to draw direct conclusions.

using an nf200 or plx on x58 will mean instead of 16/16 you can double one of the 16 links and get 16/16/16.
to get 16/16/16/16 you would need two chips.