>  > ccd0 is striped across 3 RZ29B disks on a NCR53c860 fast, narrow controller.
>  > sd0 is an IBM DDRS-39130W disk on an Qlogic 1040B controller.
>  > wd0 is an IBM-DJNA-370910 disk on the builtin IDE controller.
>  > 
>  > When reading directly from /dev/rwd0c I get slightly better numbers,
>  > 15-16MB/s sustained.  Generating a file (from /dev/zero) on wd0g
>  > writes approximately 14-15 MB/s through the file system (plain FFS).

> Modern IDE can be a bit faster than SCSI because the per-command overhead
> is lower in the protocol.  So, this isn't surprising.

I would expect performance for both IDE and SCSI to be limited by 
the rotational speeds. 

Assuming that the raw transfer rate of the two disks listed above 
are the same, where on the disk is wd0g located and/or are the writes 
going to contiguous sectors? 

Reading a disk via /dev/xxx will generally start at the outside and 
move in.  There are more sectors per revolution on the outside - 
at least on modern disks. 

The IBM-DJNA-372200 that I'm using transfers at almost 17 megabytes 
on the outside and drops to about 11 on the inside.