When constructing the usual MCV (a 3-tier architecture) one can decide to implement the MCV with double-deck interfaces, such that one can in fact replace a particular tier without having to modify even one line of code.
We often see the benefits of this, for instance in scenarios where you want to be able to use more than one database (in which case you have a double-interface between the control and data-layers).
When you put it on the View-layer (presentation), then you can (hold on!!) replace the USER interface with another machine, thereby automate REAL input (!!!) - and you can thereby run tedious usability tests thousands of times without any user having to tap and re-tap and re-re-tap the same things over and over again.
Some describe such 3-tier architecture with 1 or 2 double-interfaces as 4-tier or 5-tier architecture, implicitly implying the double-interfaces.
Other cases include (but are not limited to) the fact that you - in case of semi-or-fully replicated database-systems would practically be able to consider one of the databases as the "master", and thereby you would have a tier comprising of the master and another comprising of the slave database.
Mobile example
Therefore, multi-tier - or N-tier - indeed has a few interpretations, whereas I would surely stick to the 3-tier + extra tiers comprising of thin interface-disks wedged in between to enable said tier-swaps, and in terms of testing (particularly used on mobile devices), you can now run user tests on the real software, by simulating a users tapping in ways which the control logic cannot distinguish from a real user tapping. This is almost paramount in simulating real user tests, in that you can record all inputs from the users OTA, and then re-use the same input when doing regression tests.