the reason for the "extent size should be multiple of (block size*db_file_multiblock_read_count)" statement is that when oracle and the o/s combine to read large chunks of data, the number of bytes that will be read in a single operation will be the minimum of ...

* extent size
* default block size*db_file_multiblock_read_count
* o/s maximum i/o size

So, in your case you have block size*db_file_multiblock_read_count = 1Mb (which maybe misleadingly high -- your o/s might not support a 1Mb iu/o operation). If you have a table of 2Mb you could potentially scan it in 2 i/o operations (provided the o/s will do so, of course). However if that table is divided into extents of 40kb then each i/o operation will be limited to 40kb, and it will take "25.6" (ie. 26) i/o operations to scan the table.

If you have tables that fit within a single 40kb extent then that's ok -- you'll get your single i/o to scan the entire table, but a full scan of a multi-extent table with extent size that is not an integer multiple of 1Mb (or whatever the maximum allowed by the o/s is) will suffer.