At this point, what percentage of web development can't/shouldn't be done with Meteor? It seems like everything else is obsolete in comparison. The only non-Meteor cases I can think of involve especially large scale.
Because Meteor hides much of what's really going on, it's sometimes hard to understand just how the app is working. At least that was my impression of it when I tried it about 9 months ago.
Choosing a stack on your own means more time is spent making things work, but then you know how it works. That's why I went for a MEAN stack instead of Meteor. But I'm going to try Meteor again soon!
While Meteor does do a lot for you out of the box, it doesn't really prevent you from seeing what's going on. There are some good resources for seeing what Meteor is doing behind the scenes (meteorhacks.com and eventedmind.com) if one does want to dig in.
Even at "especially large scale" there are things that can be done to support unique scale cases, including swapping out database drivers and other components that don't work as well as you might need.
Although this may be corrected in 1.0, single page applications in Javascript like this (whether in Angular or other tools) tend to underperform in SEO (Or not perform at all). Meteor has a sort of hack that handles SEO at the moment, but it's not particularly terrific.
Meteor's database support is presently focused on MongoDB exclusively. Expect relational and other data stores to have improved support in the near future. If you were starting a project that depended on a relational database (Postgres, MySQL, MS SQL, Oracle, etc.), Meteor would probably be a poor choice today. For many use cases, however, Meteor is terrific in my experience.
Basically, if you need native MySQL support (some support already exists, but it's not in core), or if you need a web-site (SEO), not web-app.
Regarding scalability, by the time your app becomes so popular to exceed thousands of concurrent users, Meteor will have improved scalability again. They've done it a few times already with Mongo oplog tailing support and whatnot. If you already have a highly popular app, it will take you time to port it to Meteor too. So scalability may seem like a problem right now, but likely won't be one in, say, half a year.
Choosing a stack on your own means more time is spent making things work, but then you know how it works. That's why I went for a MEAN stack instead of Meteor. But I'm going to try Meteor again soon!