In the script of src/util/mtd/burn_mtd there are these lines ..

- dd conv=notrunc conv=sync bs=65536 if=${linux} of=vmlinux.bin.gz.block
With this we if exist we don't trunc and fix the size of 
vmlinux.bin.gz.block, if it's more smaller than 65536 we put 0 until 
this size.

- dd conv=notrunc conv=sync bs=63k if=${linuxbios} of=linuxbios.block
The same but with 63k.

- dd conv=notrunc if=docipl of=/dev/mtd0
We write without trunc /dev/mtd0 with the source of docipl.

- dd conv=notrunc if=docipl of=/dev/mtd0 seek=1
The same but this time we skip 1 block of "bs" (i supose that we can 
remind the value of bs in the last line, then sizeof(docipl) ).

- dd conv=notrunc if=linuxbios.block of=/dev/mtd0 seek=2
Without trunc we write linuxbios.block to mtd, skipping 2 times the last 
size (sizeof(docipl) ).

- dd conv=notrunc if=vmlinux.bin.gz.block of=/dev/mtd0 seek=128
Now without trunc write vmlinux skipping 128 times the size of 
linuxbios.block.


My questions, are:

Why we write two times docipl ?
      

First of all, you are way wrong with what seek=# means. It skips # of
(output) blocks of the output device. It has nothing to do with
sizeof(some image). Please man dd for detail.

There are two copies for IPL one at offset 0 the other at offset 512B
(seek=1).

And why we skip 128 times the size of linuxbios.block ? (i supose that 
i'm wrong, any one can help me with these)

The size of linuxbios.block is 63kB plus the 2 copies of IPL == 64kB ==
128 blocks.