This is nearly never the case. In most applications (relevant to a phone), the integers fit into 32bit, and there is no gain from moving to 64bits. Had everything else remained the same, the applications would have become slower due to all pointers being larger and needing more space in the caches.
There are of course some particular computational applications that do benefit greatly from 64bit integers, but they are rare. In the case of the iDevices, the performance benefits from the migration to 64bit are due to the other improvements, most prominently the larger register file.
----------
As a software developer I can say that there is one benefit to moving to 64bits now: At the latest when you hit 2GB of RAM, the 32bit address space becomes so cramped that you have to start playing weird memory mapping games in the kernel (and yes, this starts well before 4 GB, even though that is the hard limit).
Now even if we ignore the performance impact, the lowest-level memory handling routines in the kernel are not something you want to complicate. You want the easiest, most simple code possible in order for the code to be bug-free, and the machine to be stable.
For example read this, Linus Torvalds on PAE which talks about PAE on the x86, bit is still relevant because it explains why 64bit is beneficial before you hit 4GB.
So by moving to 2GB only after/with the move to 64bits, Apple has ensured that it can keep the kernel mean and lean in a very important area, and we can be happy that our devices don't have quirky bugs or stability issues due to complex memory management code. Simplicity FTW!
----------
I'll respond to one of these misinformed posts here: While the hard-limit for address space on 32bits is 4GB, it is also true that 1GB is the largest (power-of-two) memory that you can cleanly and easily support with 32bits.
Why do we as end users care about "cleanly and easily"? Because when things get dirty they also tend to get buggy!
