I don't think it matters.
If I have to take advantage of encode on the server to write to disk, that wouldn't seemingly be any different than taking advantage of encode on an NVR to write to disk. The difference would be "more disk" being powered by having an NVR.
In the case of a small number of cameras (maybe up to four? I'm guessing), there likely would be a reasonably minimal amount of overhead added to a server. If you're running a much larger surveillance setup with a dozen or so cameras, it probably makes sense at that point to have a dedicated NVR.
Another point for consideration is whether the server actually has to do ANYTHING. If a particular camera supports writing to SD card, then it has all of the encoding components directly in it and should be able to stream exactly the same content across the wire and write it to a server. This would sort of "guarantee" that there wouldn't be any CPU consumption on the server beyond part of the normal OS process to receive a stream of data and write it to disk.
It's also possibly that a camera may be able to stream content directly to a server and have that written directly to disk with NO encoding occurring until the time of playback (would depend on the manufacturer).
These are all points to undertand when deciding on a specific camera, I'm sure.