I don't think the type of memory used on a card alone should really matter? What the memory is concerned, things should be more about the speed and amount of memory used. I think they just tended to just use whatever type of memory was the most feasible to use in terms of performance and cost.
This doesn't tell for sure that models using either type of memory weren't better or worse in some other ways though.

What are the model numbers?

Going by how SGRAM does make a difference when some Voodoo 3 cards I would say that it can be up to 10% faster at the same clock but fast clocking sdram can easily match so it really just comes down to the quality of the card and pricing ect. SGRAM allows for reads and writes in the same cycle so that is how it gets the performance advantage.

it depends on the gpu. among the cards that i have tested, sgram version of tnt is about 10% faster than sdram, but rage128gl and tnt2ultra with sgram are equally fast, if not slower than the sdram version.

i also have sgram and sdram versions of savage4: no9 with 16mb sgram, and diamond with 32mb sdram. the no9 is a bit faster as long as its 16mb ram is not used up, but they have different clock by default: no9 at 143/166, and diamond at 160/155. i was too lazy to use powerstrip to set them to same clock, and now they are all in stockpile and i don't want to take them out again.

Savage4 has block write support, therefore SGRAM will have tiny edge in depth buffered 3d rendering and potentially big one in 2d. Either way for chips as mature as Savage4 it will hardly be noticeable difference.

My savage4 with 16mb sgram is faster than other savage cards with sdram and same clocks. My no9 savage 4 extreme is 160/166 by default