I think they consider the slow-512-writes to be a feature, in that you can run multiples of those writes and they will all perform the same, and not interfere with other transfers.
But I don't think that is the issue we talk about, your slowdown is something that happens after it has been running for a while, which might be a unique problem to osx we would need to solve.
We do have some changing coming up though, it is possible that the taskq fixes that rottegift is working on might help a little, hopefully he'll make a PR for that soon.