I think you misread, the delays are if you don't use instant delivery. I use it and it's extremely consistently delivered instantly, which makes sense, it's a websocket.
As to battery drain, I'm sure it technically does consume more, but according to my phone it's an insignificant amount: <1% of usage which is the lowest stat it gives you. Their docs suggest the same thing:
> the app has to maintain a constant connection to the server, which consumes about 0-1% of battery in 17h of use (on my phone). There has been a ton of testing and improvement around this. I think it's pretty decent now.
Honestly it's a good solution that works well with few downsides, the only real one is that iOS doesn't support doing it, but personally I don't have any apple phones so I do get an essentially free lunch.
Google doesn't have any magic way to do instant notification that nobody else has access to. The only thing they have access to in this regard is disabling any battery optimisations without triggering warnings.
Notification and battery performance is on par with google's solution except when an android build does dumb things to prevent the background activity, in which case notification performance gets worse and battery draw gets worse (not sure why exactly, it's just a common issue in these regards).
Well, there is an advantage, if everything is using the one service then you only need to have one thing alive to check it, so each new app is "free" if you already have push enabled (assuming that push notifications are rare enough the activity isn't the cost), as where each app doing it themselves is going to cause more battery use, so it isn't directly equivalent.
However, it also isn't a big deal, at least in my experience, at least for ntfy.sh.
Listening on a socket doesn't drain any battery when no data arrives unless the app does other things that actually use CPU. That's just what Google/Apple want you to believe so you depend on their proprietary lock in services.
Also like, how else would the Google / Apple services do it? Probably via sockets right? I guess you could do it in a pull-based approach on a timer, but that doesn't seem more efficient to me.
A single process waiting on multiple sockets is basically no more expensive than a single socket, but if each app has its own background process then that is more expensive. So for best performance you really want to delegate all the push-notification-listening for all the apps on a device to a single background process owned by the OS, but it'd be fine for each app to use its own push server (though of course most apps do not actually want to self-host this).
