I have an NFSv3 server running Slackware 11, and exporting directories with
the SGID bit set, like so:

/opt/dir1 /16(rw,sync,no_subtree_check)

and /opt/dir1 is set 2770:

user@nfs-server:/opt/dir1$ ls -ald
drwxrws--- 412 root groupname 12784 Apr 26 16:43 .

Now, sometime around 2.6.12 this broke, in that on the client side dir1
is mounted but a user with "groupname" as a supplementary GID cannot cd into
it. If that user makes "groupname" their effective (or primary) GID with sg,
then all is well. I remember reading somewhere that 2.6.21-rc fixed
a problem like this, where users with more than 16 supplementary groups had
problems, and that certainly applies to this setup--but 2.6.21 neither fixes
the issue nor behaves any differently than prior releases.

Has anyone else seen this issue, or is it in userland? chmodding dir1 2775
allows our client users access to the filesystem, but they still can't
create new files unless they sg to "groupname".

Clients are typically mounting the shares like so:

:/opt/dir1 /opt/dir1 nfs tcp,rw,hard,intr,nodev,rsize=16384,wsize=16384 0 0