Question if there is a known race condition for no-cache headers when using AG ?

We recently did an in-place upgrade of our our NAM 4.1 site and the Access Gateways to version
The upgrade itself went without a glitch, but afterwards we do see some strange behavior on sites running through the AG's for authentication

Some of the sessions get the access manager "resource not available" page, other work like a charm.
On our F5 firewalls we see messages that sometimes the maximum header count of 64 is exceeded, this seems to cohere with the errors.
When we set a maximum of 128 headers this setting is also met by some of the sessions.

I did a capture of the traffic with tcpdump and after decryption it shows that the no-cache header is present a _lot_ of times in the same request.
I did come across the bug where the headers seem to get corrupted by adding a #015 (https://www.netiq.com/support/kb/doc.php?id=7021103)
but that does not seem to be happening here or it's not visible in the dump.