I'm guessing it's about the rm -rf /etc/iscsi/send_targets/initiator/. In my case tis is far from empty.
In my experience there is no need for step 3) and 4) but you do need to do 1) and 2) on all nodes.
Running a remote PBS for several clusters i'm getting the impression these problems became quite rare after improving disk IO/zfs performance on the PBS. Network speed seems less relevant.
You should be able to just remove cache devices. If that makes the server hang it would probably be a hardware problem. Or maybe it was a bug that's long gone now.
I ran into this when a second node that had no local-lvm ( zfs only) was added. Adding this to storage.cfg on pve1 fixed it:
lvmthin: local-lvm
thinpool data
vgname pve
content rootdir,images
nodes pve1