In a traditional sense, N-tier means separating the application into "tiers" and putting each "tier" on different servers. This was done for at least 3 reasons:
Maintenance:
a) Code Maintenance: Easier to do bug fixes and feature additions.
b) Hardware Maintenance: Taking one server down does not disrupt service from other tier.
Performance: One server was often not fast enough to handle web requests, business logic computations, and database/file access at the same time.
Scalability: Specifically horizontal scalability
a) Fault Tolerance: Ability to have more than 1 physical server per tier means when 1 server is down, the application can still function as a whole.
b) Load-balancing: Having multiple instances of a tier helps service large amount of requests.
Nowadays, hardware and networks are fast enough to serve thousands of requests per second on a single server. Also, the buzz word for IT right now is "consolidation". So even if the application is split into tiers, they probably will end up hosting on virtual machines on a single server.
I think nowadays when people talk about N-tier architecture, they are talking about separation of concerns within the application. It is more of a logic separation than a physical one. I think as long as we achieve good separation of concerns and loose coupling, applications do not have to be N-tier. It just seems that many programmers think that N-tier architecture is a golden standard that every web application must comply with.
So, what is N-tier architecture to you nowadays?