OK, it looks like there are some limitations in the Loudness implementation I was not aware of, and I had misunderstood the Internal Volume Reference Level.
I had been under the impression that Internal Volume Reference Level was set in decibels - so you were telling Media Center that 100% Volume = xx Reference Level (dB)
It appears that the reference level is fixed at 83 dB, and must be set using -20dB narrowband pink noise. The Reference Level option is to tell Media Center where the Internal Volume Control has to be set for your system to measure 83dB. In most cases you probably want this at 100%, unless you are using a power amplifier, or need to go louder.
That said, if you are intending on using Volume Leveling, I would still set your reference level using pink noise which has been analyzed and leveled. This way all tracks are being normalized to the 83dB reference level.
Another issue is that Loudness does not seem to take effect if you increase volume above 83dB - only when you reduce it below that level.
Volume Leveling should not apply Loudness. Volume Leveling is used to bring tracks in line with the reference level - those adjustments should not have Loudness applied.
Loudness should only be applied when you are reducing your playback level below 83 dB. This needs to be done via the Internal Volume Control, and not your amplifier - once you have calibrated your system to measure 83dB and enabled Loudness, you should not touch your amplifier's volume control.