Right! When my Pi died, my network didn't look for a backup DNS, so everything became inaccessible. It was weird - probably the classic SD card issue. With NextDNS, while I do use DNS over TLS, if my Synology fails, it just kicks back to regular NextDNS domain name servers.
As long as the link doesn't drop or reorder packets too badly a >300MB/s (or probably given we're talking about small virtual machines 300Mb/s) flow doesn't require much tuning. For a single long lived TCP connection allowing the window size to scale to ~2x the expected bandwidth delay product should be enough. So a simple things like large HTTP uploads or downloads will work fine once the congestion window has grown.
The real problem is going to be anything more complex than that. Most request-response protocols too chatty and won't send enough independent outstanding requests to fill up the bandwidth delay product of an intercontinental link which will kill the throughput. Also users don't like to wait for slow responses no matter what throughput you can sustain for large transfers.