As the whole Apache tribe knows every single weblogic server in the cluster a session stickiness wouldn't make sense to me too.
I didn't find any definite answers, but I guess a simple test would be to shutdown one (or all but one) OHS while you are running several forms sessions on several nodes. If the forms sessions running on a node where you shutdown OHS survives this I guess you should have proof that session stickiness isn't necessary.
Thanks, Christian - that is a good suggestion. The reason I'm asking is that we are getting seemingly sporadic errors that the Forms session has tried to migrate to a different server (which we all know is not possible). We do have a case open with Support, but the documentation is contradicting what they are telling me, so I thought I'd ask for some more insight.
One option that I'm considering is that maybe the timeout in mod_wl_ohs is set too small, and Apache is trying to fail it over when the response doesn't come from WLS fast enough. I remember when OC4J was the bees knees, there was a nifty tool (I think it was called RHAT, I cannot remember) that would show you a graphic view of the topology of a cluster, and you could verify how things were being routed. Apart from setting the DEBUG parameter for mod_wl_ohs and digging through logs, I don't know of a good way of doing this in 11gR2.
Edit: it was called iHat
Without having a full understanding of the topology I can just make a high level guess. From the client point of view, all requests sent from CLIENT-1 must always return to the machine and server (OHS, WLS, etc) from where it originated. So if you have a LB in front of several machines that host a full installation (OHS, WLS, etc) then I would expect to need session binding. Having said this, as I mentioned, without fully understanding the layout, there may be cases where this might not be needed. Given the error(s) you mentioned, I suspect you have already answered your question and need to enable session binding. I would also recommend ensuring you have the latest WLS patches installed as I recall several issues related to session handling that were corrected. Consider these two WLS patches for 10.3.6: Patch ID 17071663 & Patch ID 13972486
Contact Oracle Support for questions about these and other patches.