Based on your setup, the inconsistency on the latency that you're experiencing possibly points toward a routing or proxy behavior difference between the external Application Load Balancer and the Classic version, rather than just a misconfiguration on your end. Though both load balancers function in Premium Tier and utilizes Google's global backbone for low-latency anycast routing through GFEs, their internal architecture are not exactly the same. For an instance, your External Load Balancer's Envoy layer with its dynamic default load balancing algorithm may re-route using alternative GFEs during intercontinental hops (for example, your test of Asia to Europe) when minor congestion occurs, which explains the 260ms-1000ms fluctuations. Meanwhile, the Classic Load Balancer sticks to a simpler, single-optimized path, minimizing fluctuations thus the consistent RTT from Seoul to europe-west2.

It might also be worth getting Google Cloud Support with all your findings to identify if this is related to a larger network problem or internal routing issue.