Okay! Here's my reasoning. I hope I'm not falling into one of those counter-intuitive
traps, so you'll have to check my maths.
The data I used for the basis of these calculations I got from the Dpreview site, ie. the D60 sensor has 135 pixels per mm, the 1Ds sensor 113.5 pixels per mm and the SD9/10 sensor 109.5 pixels per mm. (I'm referring here to full pixels containing red, green and blue elements whether interpolated or not.)
Measuring the diameter of the moon on my shot with D60/400mm combination, I find it's 3.78mm on the sensor, which translates to 510 pixels in diameter on the D60, 429 pixels on the 1Ds and 414 pixels on the SD10 sensor (using the correct terminology describing the SD10 as a 3.43MP camera).
Since it's been established that the SD9 outresolves the D60 by a noticeable margin, we can make the satement that one SD9/10 pixel has the value of two D60 pixels for any given area of sensor.
Since it's been established, on this forum at great length, that the Canon 10D does indeed exhibit higher resolution than the 1Ds when using identical lenses on both cameras, I think it's a fair assumption to make that one SD10 pixel is also worth two 1Ds pixels.
To normalise the figures for all three cameras, we get 510 pixels across the moon diameter with the D60, 429 pixels with the 1Ds and 414 x (square root of 2) = 580 pixels with the SD10.
Now I'm sure everyone would be prepared to accept that a moon that's 580 pixels wide will have significantly higher resolution that the same moon that's only 429 pixels wide.
Okay! 400mm is not particularly long. I believe the longest lens that would fit all three cameras would be the Sigma 800mm. The moon now becomes 7.56mm in diameter on all three sensors. Let's add a 2x extender. It becomes 15.12mm in diameter. Oops! The SD10 is now cropping the image and the D60 is borderline. (However, if it isn't a full moon, you'd probably still fit it in the frame
Even if we compare the 1Ds/800mm/2x combination with the SD10/800mm/1.4x combination, we're comparing 1427 SD10 pixels with 1716 1Ds pixels. The images might be on a par, but it's doubtful. A 1.4x extender at best offers a marginal improvement. A 2x extender is even more marginal. Those extra pixels in the 1Ds image would not be doing much.
I rest my case. Thanks for giving me the opportunity to exercise my mind with some basic maths, and I'm sure you'll point out any errors in my reasoning