Indeed. To get /proc to work over NFS you have to give the files their
correct length, and that is not done by the normal kernel (because the length
calculations are non-trivial).
What you _can_ do to fake it and to generally make it work for most cases is
to just create a patched kernel that fakes the lengths to some random value.
I've used this once to debug a kernel problem over NFS. Sick, but potentially
useful (I made all /proc files be 4096 bytes in length, which happens to be a
nice round random number).
Linus