Anyone else having issues with GCP's internal HTTP load balancer and WebSockets?

Ran a 24-hour WebSocket stress test across GCP, AWS, Azure. GCP's internal HTTP(S) LB is the outlier.

* **Test setup:** 500 persistent WS connections, 1KB messages every 5 seconds. Backend: GKE pods (Nginx as WS proxy).
* **GCP ILB Config:**
```yaml
apiVersion: networking.gke.io/v1
kind: FrontendConfig
metadata:
name: websocket-config
spec:
timeoutSec: 3600
```
* **Issue:** Connections drop between 300-350 seconds, regardless of `timeoutSec` setting. No errors in LB logs. Backend shows RST_STREAM.
* AWS NLB & Azure Internal LB held connections for full duration.

Is this a known config gap, or is the 300s timeout a hard limit for WS on GCP's internal HTTP LB?