32-bit vs 64-bit
There's a lot of confusion about 64-bit versus 32-bit processors. What is generally meant by this is the size of a general-purpose register or a memory address. There are lots of other bit measurements -- size of data bus, size of floating-pointer registers, special-purpose registers (e.g. Altivec), etc.
For _most_ applications, you won't see any speed improvement (and possibly a slight penalty) from 64-bit mode. Why? Very few applications need to do integer math on numbers larger than 2**32 (~ 4 billion). Very few applications need memory addressing beyond 4GB. Very few systems have hardware RAM beyond 4GB. So, in essence, you're pushing around twice the data every time you work with a memory address or an integer.
However, for large-memory applications (databases are a good example), having >4GB addressibility is a _huge_ win. Why? They don't have to swap pieces of data in and out, or if they do they can let the OS handle it. Less pointer manipulation, look-up tables, etc.
Also, in the not-too-distant future we'll see home machines with >4GB RAM, at the rate we're going. A 32-bit system could make use of it through segmenting the address space, but no application could address any more than 4GB. In fact, the kernel itself couldn't address more than 4GB at a time -- it'd have to change which 4GB of physical RAM mapped into its 4GB of addressible space.
All of the above is why most 64-bit processors maintain a 32-bit mode as well. 32-bit applications run without modification (and without penalty) while 64-bit applications can enjoy the extra space. The kernel itself can run in 32-bit mode while some applications are 64-bit (AIX does this; I believe others may too). In fact, if you look at AIX (which has supported 64-bit applications since 4.3 and 64-bit kernel mode since 5.0), most apps are still 32-bit. Only the few that need the extra space are 64-bit.
The reason why the 970 should be a big step forward is only partly for 64-bit -- until >4GB desktops are common, and apps are ported to use the address space, it's not a big deal. There's room to use it for video work, but it's still not a large part of Apple's business. The real improvements are a much better core, based on Power4, and likely a lot of headroom to scale up performace (as Power4 has scaled up).