Comment 4 for bug 1315955

Our workaround so far is to force all clients to connect with nfs3 instead of nfs4 and to avoid doing major operations like file tree syncs over nfs, doing them on the nfsd host itself.

Either nfs3 doesn't get stuck and crash (case 1) or we're just lucky it hasn't returned yet, with minor mitigation by dropping load (case 2).