Below is a copy of a message of mine to another list a few months back - which lists some basic real world results for two _similar_ machines that I use each day.
Remember that these results mean nothing to anyone except me - but it can be interesting to look into these sorts of things.
Basically - if your OS or application can use DP, then you will get a speed increase, but nothing near 200% or even 150% in most cases.
P.S. The benchmarking action I used did not use similar settings to my day to day production - so the time/speed may be a bit misleading (there can be a huge jump in time by using a 2 pix vs a 30 pix blur etc).
======
Unit A: G4 400, Single Processor, 370 Photoshop RAM allocation
Unit B: G4 450 Dual Processor, 230 Photoshop RAM allocation
After running a benchmarking action on both machines set to the same
nearest possible test conditions of each computer being at it's optimum
'real world' performance (chock full of apps and files and not defragged,
but with more than enough scratch for the tasks at hand):
50mb RGB Test File -
3.7 pix G/Blur:
Unit A: 25.6 sec
Unit B: 12.7 sec
85 pix G/Blur:
Unit A: 28.2 sec
Unit B: 13.6 sec
50/1/0 USM
Unit A: 12.7 sec
Unit B: 8.8 sec
Despeckle:
Unit A: 12.4 sec
Unit B: 8.5 sec
RGB to CMYK:
Unit A: 39.0 sec
Unit B: 37.3 sec
60% Image Size Width
Unit A: 4.2 sec
Unit B: 4.0 sec
There are more results and more tests to do (this is only part of the
benchmarking actions results), but it is obvious that even with less RAM
the dual processor running 50mhz faster will speed up many of my
operations (no real shock there).
So that's it - there's no real point to this post, just sharing my initial
findings on two real world computers (as in the past I have only had
reviews to go by and these are usually stacked or not real world setups).
This is not designed to test SP vs DP or I would have set both RAM the
same...it is just an attempt to see which of the two boxes I have access
to would be better to use as the main box - unless I am missing something
big it would seem that the dual processor is the winner for my work (which
is 99% Photoshop).
I found this interesting link while doing a search on MP Photoshop
http://www.reed.edu/~cosmo/pt/tips/Multi.html
It seems that the benchmarking action is not 'ideal' for me, since many of
it's tests are not 'real world' for my tasks - but it does provide some
good answers anyway. Can't recall where I got it from, probably a search
of the web for Photoshop benchmarking action or something.
I rasterize many Quark pages (selected elements) into Photoshop so that I
can perform seamless soft masking between elements that are usually alien
to each other (vector/raster).
Rasterizing a single or double page spread into Photoshop can be time
consuming, as the resolutions are often higher than normal - since the
Quark elements need to stay sharp (but 300 ppi is often acceptable), these
are panels and perhaps headline type - nothing really sharp or small, that
is left as vector in layout.
Rasterize 28 mb Quark EPS single page to 300 ppi CMYK:
Single processor - 71 seconds (half of this time is with the progress bar
stuck on 0%...then suddenly it churns through the data)
Dual processor - 44 seconds (instant rasterization progress with no stall)
Stephen Marsh.