Well, the PERC 5/i "sucked" for mirrors but from what I read it'd probably do fine with RAID 5. I wasn't able to test it in that configuration.
Honestly what's probably killing you on the old server are the 7200 RPM drives. We're moving from a 3Ware-based mirror and, well, you can see that in my iometer results under the "current" tab. Those were 7200 RPM drives as well. With the "SATA Database" graph you see the SmartArray P800 at top, then in order it's the Windows-based mirror with 2 7200 RPM drives running on the PERC 5/i, then the PERC-based mirror, then a 3Ware 9500 mirror, then a regular SATA controller with a single drive, and finally a regular SATA controller with Windows RAID. Yeah, I dunno why the single drive scaled better than SATA windows mirror, but there ya have it.
If I had any spare 15k drives, I could run some more tests on that PERC 5/i, but that system is set up to take 3.5" SAS drives and all my 15k ones are 2.5" SAS so it's kinda ghetto trying to prop them up.
You still may want to investigate getting a proper RAID controller, though, be it a PERC H700 or older PERC 6/i or something like that MegaRAID as even though the H200 is faster than your old system, you'll probably be leaving performance on the table with the SSDs.
At the very least, once you have a benchmark routine down, you could buy a controller from some place with a decent return policy and see if a proper battery-backed cached controller gives you any real-world performance. If it don't, return it. That's part of the reason I wanted to get iometer working for me so I could figure out if I were heading in the right direction with hardware and tuning for the expected workloads on the servers I was buildin'.
More PERC blah blah blah: http://www.dell.com/downloads/global/pr ... debook.pdf
BTW, what was the RAID controller in the PE840 you tested? And did it have cache and a working battery?