I don't think that there's any evidence that the linux swapfile performs better - and in any case why would it being unfragmented be an advantage? Memory access is random, and so swapfile access is random, and so why does having it non-contiguous cause an issue? Added to which, SSDs are becoming much more widespread, meaning the fragmentation issue vanishes.
In any case, can you make Linux use a swapfile permanently? One system I look after needs more swap, and I really don't want to repartition the entire drive just to increase the available virtual memory.
dphys-swapfile is available on all distributions that I know of.
Still, swapping from a partition will be faster than swapping from a file. There are simply less layers to traverse, but whether it's a perceptible difference, I just don't know as I've not measured it.
But you're right, SSDs are obviating the need for a contiguous swap file/partition. Swapping is inherently random in nature, which translates to random IO on contiguous or non-contiguous swap.
I have wondered what sort of difference log structured swapping would have? Swapping out would reduce to a contiguous operation as you'd be writing to the end of the log, and swapping in is effectively random IO anyway, so would not be impacted by the fragmentation introduced by the log structured writing. Again, not that much of an issue with SSD, but with HDD, it might be a measurable benefit when trying to free memory.