> Once I realized they were over clocking the processor it occurred to me that
> they were probably not using an ideal ratio to derive their memory clock (that
> they are actually running memory slower than it could be). From the new
> figures you linked, I'd say they've figured out a way to up the memory clock

"On the Sam460ex the DDR is overclocked as the CPU."

That means both the CPU and the memory bus run overclocked by 15%. If I'm not mistaken this would make for a memory clock of 230 MHz. Whether that's really the reason for the reported improvement I can't say. The old figures had been reported at a CPU clock of 1.167 GHz (i.e. overclocked by 16.7%), which presumably translates to a memory clock of either 233 MHz (5:1 ratio) or 212 MHz (5.5:1 ratio, which would fit your theory of increased memory clock).

> or make some other change - like lowering latency

This would be the case where they kept the 5:1 ratio, i.e. actually lowered the memory clock from 233 to 230 MHz.

> There may even be some additional tweaks they could make.

Yes, that's what is said:

"Next U-Boot versions for our board will include modifications on the DDR config code to achive higher speed transfers."

> I wonder how reliable this system will be if they're resorting to these tricks?

"On the Sam440ep models, the DDR runs [...] overclocked on the 733 Mhz model (147 Mhz)"

Apparently, the overclocking of the Sam460ex is similar to the 733 MHz Sam440ep-flex where both the CPU and the memory bus run overclocked by 10%. So far, I didn't hear about reliability problems on this machine.