I've been doing a more "scientific" test, trying to get more repeatable results, and have come to some interesting conclusions that are a little different to what I first thought. I used a ballbearing rolled off a ramp to ensure a consistent speed when cutting the beam - I took several shots with each setup and the results were very consistent.

Firstly, there are significant differences between teh 5D and the 1DII in these tests, both in absolute terms and in behaviour.

Secondly, there are more variables than simply MLU vs NOT - whether the circuit is arranged to half-press the shutter ahead of time makes a lot of difference. In particular, if the shutter is not half-pressed ahead of time, there is a substantial initial delay (40 ms or so). If the button is released during this time then no shot is taken. All tests I did on this session were taken starting with the button half-pressed, manual focus, lens wide open, manual exposure.

On the 5D, MLU makes a significant difference to the lag

Irritatingly, on the 5D with MLU enabled, the shutter button needs to be COMPLETELY released (i.e. the half-press circuit needs to be disconnected too) after the initial press that locks up the mirror, or the second press to take the shot is not recognised.

In the 1dII, the lag is a lot smaller than the 5d, but seems to be more-or-less unaffected by MLU. In fact, it got slightly longer! The MLU does not require the button to be completely released on the first press.

The personal function on the 1DII to reduce the lag where possible did not seem to make any difference.

The shot in the middle is using the multi-flash feature of the 580EX set at 100 hz (ie a flash every 10 milliseconds)