Well, that's a good point.

Inspection of the board reveals a 32.768 kHz crystal at location
6F. This is the crystal used to keep time in the system: 2^15 = 32768,
which is a nice easy number to divide down to get a precisely 1 second
pulse with a 15-bit counter. 

This crystal is hooked up to the MCPX. This, perhaps, answers one
question: the backup voltage is 2.5V because the time keeper is in the
MCPX, which is a 0.13u part and it is possible that its transistors
can't deal with a 3V backup battery. Nonetheless, I disconnected the +
terminal of the capacitor and measured the leakage current with the
power off: this is 0.14mA. Note that if you totally disconnect the
capacitor and the power, the current goes way down, but I presume
that's because the clock shuts itself off.

Q=CV, so the capacitor holds 2.5 couloumbs of charge when fully
charged. At a discharge rate of 0.14 mC/s, the capacitor should last
about 5 hours to zero, although the clock probably stops working
around 1.3V, so this confirms xboxmagic's 2-3 hour number.

I discovered that why my clock sometimes only stays on for about 5
minutes is that it takes a minute or two for the capacitor to fully
charge. (There is some series resistor of low value that prevents a
*huge* current spike from rushing in if it's discharged). Because I
tend to errr...power on my box sporadically for short periods err...I
have "odd" gaming habits... ...the capacitor never gets fully charged
and thus it looks like it's not keeping time.

A 0.14 mA draw is enormous for a battery-backed clock; I looked up the
Ibatt for one RTC and it's like 300 nA...about a factor of 500
less. The MCPX probably draws so much current because it's done in a
.13u process, and the leakage current at those gate sizes is horrible.

So, battery backup? well, even if you used a CR2032 coin lithium
battery (220 mAh capacity) with a diode in series to drop the voltage
down to 2.5V, it would last probably less than 60 days :/ Plus, in
order to drop a battery in, you'd have to disconnect the capacitor
charging switch.

So in the end, the answer is: MSFT cut cost by not putting a long-life
real time clock part on the XBOX, and instead integrated it into the
MCPX; because of this, the current draw was high and they were forced
to use a supercap which charges while powered on, becuase if they
didn't by the time the box was shipped to the customer, the battery
would be mostly dead.

Eh. back to thesis.