>We use STUN-style discovery and relay fallback

How does your relay compare to Tailscale's (DERP)?