Inet port exhaustion: no virtual involved

We opened a support case because ltm logs were showing Inet port exhaustion. No virtual is involved: non-floating self-ip sending UDP syslog traffic (iRule for w3c logging from devcentral).

I found the cause to be at least indirectly related to the self-ip's VLAN having cmp-hash set to source-ip instead of default. Has anyone else seen this? Switch vlan to default and problem goes away completely; repeatable. There is very little traffic going to the syslog server - 5 lines per second peak. Tcpdump during reported port exhaustion shows nothing unusual.

The exhaustion is happening on a Viprion with a single B2250 blade running 11.5.1 HF10. I'd like to find a workaround because we'll need cmp-hash set to source-ip for effective bandwidth controller profiles.