I've never heard any claims or evidence that 32/64 bit ODs have
any
impact on the harddisk access times. Someone please correct me if I'm
wrong but this is not the reason people use 64 bit OSes.
None at all on a typical system. Disk transfers are 16bit for ATA anyway
(32bit PIO as a special case we don't normally do but some day I should
add for the drives that do it), and DMA is used for all real disks today.
As far as I am aware there are two reasons.
1. For technical reasons, a 64 bit build is normally a bit faster than a
32 bit build, just due to compiler options. For normal desktop usages
its unlikely you would notice a difference, but it can have one if you
do heavy number crunching.
Its a bit more complicated. Generally speaking on systems user 64bit
binaries are slower than 32bit as they are bigger. PC (x86/x86-64) is a
bit different as x86-64 is very compact and also because it doubles the
number of usable registers in the instruction set which is a huge win. So
typically its about 10-20% faster. Desktop performance is usually bounded
by video performance (lots of opportunity for the X folks to improve
performance left) and most of the time the box is idle anyway so being
able to display an icon 20% faster isn't a big visible hit.
2. 64 bit OSes, have a larger memory address they can access. With 32bit
OSes the largest amount of memory a single process can use is 4G I
think. With 64bit this problem goes away. You may wonder why any single
process may need that much ram... Its true not many will but some tasks,
like large databases or image processing can get that high.
Even more importantly - the kernel can address more than 4GB at a time
without doing expensive memory management unit updates. That makes a very
big difference above 4GB and quite a big one above 1GB RAM (the kernel
normally has 1GB of RAM and 3GB available for the user application - more
requires switching between the user and physical memory view which is not
cheap)