smart array read cache algorithm

Hello.

One of our customer have :
dl360 g3.
smart array 5312 256 cache.
linux AS 2.1
enclosure 4354r.
6 146gb disks in raid5 - chunk size 16k.

He is doing dd to /dev/null and when he using

64k block size the sector/sec droped after few
sec from 130000 to 66000. if we disabled the 5312 read cache the performance rise again to 130000. , With bigger block
size like 320 512 640k he get about 170000
sector/sec no matter the read cache enabled or
disabled.

He wants answers From HP why This happen & I
Dont have it.

Can someone know if its bug in the read cache
Algorithm or its should as we notice in our tests.

Thanks


Shalom.