What's faster, qemu-nbd or vdfuse? Which support differencing images (snapshots)?
I suspect fuse to be slow, but nbd seems to be some old stuff.
What makes you think that because it's old, it's not what you need? NBD has been around for a while, but is still being actively maintained.
In theory (I haven't actually tested any of this), since NBD works at the block layer, the kernel should be able to cache more, which should allow it to work faster. The downside however is that it is not possible to access the same device from more that one client machine; in contrast, there is nothing inherent to FUSE which would have the same limitations. In practice, whether that's the case depends on the actual way in which it's been implemented.
(disclaimer: I'm one of those who help out with maintaining NBD)