I have done some timings myself on a P133 Machine (160MB RAM, Tekram DC390)
running Linux 2.2.0.
Here are the results:
|Software||24 bpp RGB at 2700 dpi||24 bpp RGB at 1350 dpi||32 bpp RGBI at 2700 dpi||32 bpp RGBI at 1350 dpi|
|Linux / xscanimage / sane 1.0.1 /coolscan 0.4.1 - no data transfer||25 sec.||25 sec||42 sec.||40 sec|
|Linux / xscanimage / sane 1.0.1 /coolscan 0.4.1 - including data transfer||50 sec.||30 sec||1:45||55 sec|
|Software||24 bpp RGB at 2700 dpi||24 bpp RGB at 1350 dpi||40 bpp RGBI at 2700 dpi||40 bpp RGBI at 1350 dpi|
|Win95 / Nikon Scan / TWAIN /incl. dust removal||2:15||35 sec||6:40||1:40|
|Win95 / Vuescan / scanning only / no saving / 80 MB buffer||25 sec||25 sec||48 sec||48 sec|
The last two columns are scanning times for reading RGBI images (including the infrared channel). Under SANE and vuescan, only the scanning times are measured, not the actual dust removal. Under TWAIN the scanning and dust removal is measured together (as there is no option to seperate them).
The TWAIN driver sacrifices a lot of performance by mixing scanning and image processing tasks.
If you have done similar measurements please send me the results so we can see how machine dependent this is.