[tracker1]

There's an additional aspect... Types of load (compute vs i/o)... I'm currently building a system where node/iojs is the primary application runtime. In this case, separating those portions of the application that are compute heavy (resizing images, recalculating values, etc) are offloaded, beyond that I have separated the backend data into discrete containers based on collections of data/types... that which is related to an account, vs that which is related to a user/profile (and authentication)... which makes a bit more sense. I also separate the authorization as a service layer above backing logic.

UI (shared client/server codebase), WebAPI (ui access, auth filtering), DataAPI (used from WebAPI), Profile Service (users/auth), Account Service (core data), Image Service (backs upload/resize) and various backing pieces... for me I didn't break things up into too fine grained pieces, but wanted breaks based on type of work and/or data.

It really just depends on your workload.