Certainly I also can't imagine how you would get strong performance out of Server Blazor in real world/Enterprise conditions. RPC busses are going to RPC and roundtrips are still roundtrips even when SignalR-backed web sockets. Flooding UDP web sockets with HTML chunks isn't necessarily a great idea, even if HTTP3 (QUIC) kind of/mostly succeeds at it, but SignalR as good as it is certainly isn't HTTP3.

On the other hand, Client Blazor has huge initial overhead (it compiles a form of the entire .NET platform in WASM; you can golf it down a bit, but it's still an entire runtime and GC and most of a standard lib depending on what your app does, and its dependencies), but once it is loaded it can be as performant as just about any other Virtual DOM-like (modulo the overhead between DOM/JS and WASM, but that overhead is surprisingly small in current browsers and seems to only get smaller as browsers continue to optimize WASM scenarios).

I don't think I'd recommend it for most consumer-facing websites any time soon, but in my experience Client Blazor is still much better than Angular for performance in an Enterprise/Internal App scenario.