Wellll....first of all, you should *NEVER* use any other io scheduler in Linux other than noop for any DBMS device - doesn't matter which DBMS. So that could be part of the problem. The other part may be the IO request queue size for the block devices:
cat /sys/block/<device>/queue/nr_requests
If less than 1024, set to 1024 via:
echo 1024 > /sys/block/<device>/queue/nr_requests
Unfortunately, this isn't persistent, so you will need to add it to rc.local. The other thing to check is what filesystem settings are being used. Ideally, you should run without journaling - preferably with xfs or ext4 (or best case Veritas vxfs). To turn off journaling, you would use:
tune2fs ^has_journal
....I mention journaling because you point out that read performance is better, but write performance is worse - and one of the leading causes of slow writes is journaling. Journaling is one thing if you are using a lot of filesystem buffer cache - but if you are using directio, it is irrelevant. You might also want to check sysdevices and make sure they are all using directio - one other possibility is that they are using dsync instead. But back to the filesystem, you also should check the mount options
noatime,nodiratime
should help a little (very little, but every bit helps). If your sysadmin is a typical OS admin with no real DBMS background and thus playing ostrich and not letting you turn off journaling, on ext4, you would then need:
data=writeback,journal_async_commit,nobh,barrier=0
or on xfs:
noatime,nodiratime,nobarrier,logbufs=8
Once you get that setup, you might want to do some regular iostat's on both and then show the OS admin the diffs. There are VMware BIOS settings (see ASE 15.7 sp100 New Features Guides) that are fairly basic to check as well.....but remember, a VM is running along side other VM's - you may have to check the VM perf stats, but it also is possible that the machine is overloaded with VM's and therefore the ASE VM is getting bumped at times. A second thing to check is that a lot of VM based systems switched to iSCSI....but the sysadmins leave the network adapters at 'auto' - which in my experience tend to end up at 1Gbs instead of the higher 10Gbs they should operate at.....or they are using 1Gbs cards and the older HW used 4Gbs HBA's.....