First, do not fiddle with it. Right now, the data is probably still there, but every time you touch it, you risk changing that.
Second — contact your RAID vendor for recovery procedures and/or tools. For example, I've had the metadata on a 3ware (now AMCC) array get messed up — my fault, PEBKAC — and 3ware support worked with me to repair the mess, including sending tools and reconstructing the metadata. If your vendor can't/won't help you, keep that in mind next time your consider a purchase.
Now, assuming you can not get support from your vendor, I'd first make copies of each disk, block-by-block by moving them (one at a time, and make sure to label so you can put them back to the same connector) to a plain-old SATA controller, and then only work on copies of the disks. Then, with the disk copies, I'd try and see if I could get Linux's MD driver to handle it, by trying to figure out (a) the offset to the first block on each disk and (b) the block size by using a hex dumper such as xxd. Then set up loopback devices if needed (for offsets) and use mdadm --build, definitely in read-only mode! Confirm data is correct, if so, copy data. Save original disks, and maybe copies, for a while Just in Case™
(Come to think of it, there are probably tools to do that all for you)
edit
A quick search for Linux programs (because Linux will likely go around your software RAID, and be able to get to the raw disks):
- dmraid supports Silicon Image Medley software RAID, though it may be confused if the metadata is borked.
- testdisk: supposedly will scan for files, no idea if it can handle your SiI software RAID.
Also, if all three disks are readable, recovery will be easier: you won't have to use any parity blocks to read the data, and you will be able to use the parity blocks to figure the array offset/block size/etc.
Quick Googling gives a guide of how to do this: http://www.raidrecoveryguide.com/recovery.html
