Re the "banding problem"
My first reaction was that the scan is being done "off" a native
resolution 4000 dpi, 2000 dpi, 1333.333 dpi, 1000dpi etc and that software
interpolation was/is being done.
After a few of the other comments about possible mechanical
problems I remember watching either my AT210 (flatbed) or an HP
doing it's "scan dance" where it scans forward, pauses while the
programed IO SCSI interface dumps the scan buffer, backs up past
the backlash of the gears then scans forward for another chunk.
A lot of the early scanners had poor SCSI performance.
Does the scanner seem to stop and start or is it a smooth scan?
An analogy is with many SCSI tapes that are "streamers". As long
as you keep them fed with data they will keep writing (or reading) if data
stops the drive writes a "stretch mark" hoping to see more data soon, if no
write data is provided the drive stops, when you write again the drive has to
back up past the last data then read past the erased area where it starts the
next block. The stops and starts waste tape and slow down the drive, we solved
that back in the late 80's with the BSD dump routines and multiple write and
read buffers and proceses.
So is it possible that your scanner is out running your system,
the scanner stops and has to back up. It could also be a similar
problem that the data rate from the CCD head is higher than what the
Scanner interface can handle and the microcode/firmware in the
scanner is doing the "back up and scan a swath" dance.
Stephen N. Kogge