It varies widely based on where the choke-points are in the architecture, but it is usually something like:
Interweb--> Load Balancer--> Web servers--->Database Servers
The "Load Balancer" part can be as simple as a machine (or two) running haproxy, to an F5 style device, to what is effectively load-balancing-as-a-service (Akamai, etc.).
The web servers could be multiple iterations of a single server, or a pseudo CDN setup where for a given set of machines that serve a page request, 1 machine handles serving images, 1 serves javascripts, css, and other static text, and another generates all the dynamic content.
The database layer can often have its own load-balancing within it, depending on the type and scale of activity. Much of this depends on how frequently the dynamic content changes and needs to be kept up to date (a system handling comments for a busy site needs to be more real-time than a system that just updates news stories over some period of time.