Several factors may be at play. We deployed recently using an F5 but, because as part of an initiative to deprecate old SSL/TLS protocols and ciphers, we set it up in routed mode (where F5 behaves like a gateway vs. SNAT and the like) so the CAS servers themselves can directly observe protocols/ciphers in use, trap deprecated ones and display a warning page.
Anyhow, we discovered any CAS client host (configured with the CAS virtual address) on the same subnet as the CAS servers didn't work because of a layer-2 short circuit. The solution was to put the CAS servers on their own subnet (here a /28) with no other potential CAS clients on that same net.
Other than that, CAS 4.2(?) on no longer requires session stickiness. We disabled it in the F5 and see traffic pretty evenly sprayed across all the servers w/ no ill effect. E.g. host1 serves the login page, and host2 accepts the POST.
Tom.