Theoretically, performance of video output should be only limited by 
bandwidth of CPU <-> Video bus ( for PCI bus it's 133 Mb/s or 5 ms for
drawing of 640x480x16 picture ). Unfortunately, this value is rarely
achieved. Actual performance is affected by many factors and can vary 
from 20 to 250 Mb/s.
  You can determine this value for your system by running 'benchmark'
tool. Type:
    benchmark <your-AVI-file>  (try also with options -yuv and -direct)
and it'll perform quick measurement of most typical timings. In the end,
it'll print video output speed in megabytes per second. If the resulting
value for your system is higher than 50 Mb/s, then there's nothing for
you to worry about, because your X server has some kind of output
acceleration.
  In case of this performance being worse than 30 Mb/s, you'll probably
need to turn on write-combining for video card. This should generally be
done by X server, but some servers ( for example, X server for XFree86 
3.3.6 with s3_savage driver ) don't do it ( shame on them ). Here I'll
describe how it could be done by hands in case of Linux.
   1. Your processor should have MTRR registers ( Pentium Pro, Pentium II
or maybe some compatible processors ). Your kernel should be compiled 
with MTRR support: in kernel configuration menu, select 'Processor type
and features' ->'MTRR (Memory Type Range Register) support'.
   2. Run X and log its output somewhere:

 startx >./startx.log 2>&1   

  Look into the log and find a line similar to this:
  
(--) SVGA: PCI: S3 Savage4 rev 2, Memory @ 0xe5000000, 0xd8000000

  Here 0xe5000000 and 0xd8000000 are virtual addresses of your video 
memory. You'll need one of them. 
  2a. Alternatively, you can look at /proc/pci and find section related to 
your video card ( "VGA compatible controller" ). Take the first address 
from the line saying about "prefetchable 32 bit memory".

  3. Type as root:
 
  echo "base=<your-base> size=0x800000 type=write-combining" >>/proc/mtrr
  
replacing <your-base> with address received in step 2 and 0x800000 ( which
is equal to 8 Mb ) with size of your video memory.

 That's it. If you did everything properly, you should notice significant
increase in output performance. For me it increased output bandwidth from
 20 to 36 Mb/s ( 1.8x ), and I had reports about 3x..4x increase!
 
 In case that you chose wrong address from list in step 2, do the following:

 cat /proc/mtrr
 
 Find the line that was added by you. Suppose it starts with "reg02",
  then type:

 echo "disable=2" >>/proc/mtrr

 and repeat step 3 with another address.
 
 It is a good idea to put command from step 3 somewhere in startup scripts,
because that base address does not change when you restart computer, but do
not forget to remove it when changing hardware configuration.

 You can find details about using /proc/mtrr in Linux kernel sources, file
Documentation/mtrr.txt.