> Mainly to provide consistency with other tools that test file integrity 
> such as cksum(1). (Typically, when sha256 is available, sha512 also is)

Are you saying we'll increase the diversity of checks all over the place,
increasing confusion, increasing requirements?
 
> SHA-512 performs significantly faster (on 64 bit machines) when I use 
> mtree with my diff on thousands of files.

Are you saying it performs significantly poorer on other machines?

> Whilst it's not a need, it's nice to have the option.

I think nothing needs it.  I'm personally waiting for SHA1024 because it
is twice as good as SHA512.