Share your successful ReadyNAS hardware setup

The bottlenecks in storage like this are:
1. CPU overhead processing the IP stack at the NIC(s)
2. The RAM available as a cache to the above input stream
3. The CPU overhead in managing the file system and reading/writing to the disk channels.
4. The CPU overhead in hashing the data for RAID operations for reading/writing
5. The number of spindles and media data rates on the drives (drive on-board cache to media transfer speeds)

#1 is improved by offloading the CPU processing of various redundant tasks - high end NICs for servers do this, at a dollar cost.
#2 is improved by adding memory, at cost, with benefits to writes-to-NAS only, up to a point of the largest files it can cache.
#3 is improved by DMA Host bus controllers doing the read/write processing from memory. The CPU still does the file system, and so a dual CPU host is a good solution if much of the IP Stack and downstream RAID is also being processed by the CPU.
#4 Hardware RAID read/write processing on a host bus adapter is faster than having the system CPU do it in a multitasking OS. RAID rebuild is also more efficient. A battery-backed write cache module is more expensive than a read-cache-only RAID controller, and both are more expensive than doing it in main system RAM with the general purpose CPU...
#5 More disk spindles (smaller data chunks spread across more targets, transferred theoretically in parallel), higher spindle speeds (implying higher head<->platter data transfers) equals better performance if all the electronics upstream are up to the overall task.

There are many variables in the calculation, hence the tendency not to quote speeds - right now (Fall 2011), there is a shortage of hard disks due to parts made in China where factories were destroyed by various natural disasters recently. A change of memory type, hard disk maker, or other part in this could affect the overall performance, if not the price point for profitability of the model.