The problem is consistent and reproducable.  There is a workaround,
which may help locate the problem.

-----------------------------------------------------------------------
Whatever tables in the kernel that store the physical disk parameters
( C/H/S ) are not getting updated by the hdparm -b   disk swap process.
-----------------------------------------------------------------------

Start with a 200 GB drive (Maxtor 6Y200P0, 2039382109056 bytes) 

   root> sfdisk -g /dev/hdg                            
   /dev/hdg: 24792 cylinders, 255 heads, 63 sectors/track

   root> hdparm -b 0 /dev/hdg

Remove 200 GB drive, load 250 GB drive (Maxtor 6Y250P0, 251000193024 bytes)

   root> hdparm -b 1 /dev/hdg

   root> sfdisk -g /dev/hdg          
   /dev/hdg: 24792 cylinders, 255 heads, 63 sectors/track

That should be 30515 cylinders!  fdisk reports the same wrong 24792 cyl, 
which is of course the drive information for the previously inserted
and subsequently replaced 200GB drive.  hdparm -I /dev/hdg always
reports the correct information.

This affects not only sfdisk/fdisk.  Lets say I actually have a large
disk mounted, and the kernel thinks it has a smaller disk.  I get kernel
errors when I try to write to outer regions of the larger disk
(/var/log/messages):

   Dec 21 16:54:10 gate kernel: attempt to access beyond end of device
   Dec 21 16:54:10 gate kernel: 22:02: rw=0, want=188481544, limit=188418352
   Dec 21 16:54:10 gate kernel: EXT3-fs error (device ide3(34,2)): read_inode_bitmap: Cannot read inode bitmap - block_group = 1438, inode_bitmap = 47120385
   Dec 21 16:54:10 gate kernel: EXT3-fs error (device ide3(34,2)) in ext3_new_inode: IO failure
   
---------------------
The workaround:

If I do  
   root> sfdisk /dev/hdg          

... then control-C out of it (supposedly nothing changed), the kernel finds
the correct parameters for the new disk.
---------------------

I am using stock Fedora Core 1, 2.4.22-1.2115 (this also occurs with a
recompiled Win4lin version).  I normally use a Promise Ultra133 TX2
Chipset controller, but have the same problem with a Siig HPT controller.  
The same problem appears when swapping from the Maxtor 250 to a WD 200.

   /proc/ide/drivers says:

   ide-scsi version 0.93
   ide-floppy version 0.99.newide
   ide-disk version 1.18
   ide-default version 0.9.newide

... if that helps.

What other information would you like?  I can recompile kernels with
patches, though I am not very sophisticated about it, and I would rather
not get highly experimental since this is my server and my backup drives. 
I am not too worried, given the workaround, but perhaps this spotlights
a more dangerous flaw you should know about.