I think I know what may be the reason - the small number of 'white' Mars pixels and the internal threshold default for the number of pixels that Levels and Colour ignores to set the white point (or none of the Mars pixels has all 3 channels maxed out). It looks like the latter.
The dynamic range highlight tick decided to become 72.9% instead of being 100% or close, and the highlight slider matched this position when Dynamic Range was set to Reset. As a result, the highlight slider reduced the HSL-L values of the Mars pixels way down the upper HSL cone causing the maxed out pixels in some, but not all channels to get saturated.
If the white point slider is moved to the right, or set to Full, those pixels should become 'white' again.
The threshold for finding the black and white points in Levels and Color is pre-set at 0.01%, so if the area of the image covered by the white objects is less than that percentage of the total number of pixels they will be ignored and an incorrect white point will be selected. There is also something similar in the auto white balance setting.
Levels and Color is a simplified version of the Adjust transformation, and it is definitely not optimized for processing astro images. Adjust should work much better since it uses a curve instead of sliders.
If you want me to look into it further, I will need a copy of the image you are using.
Mark's image is some 16 million pixels. 0.01% of that would be 1600 brightest pixels that, if I understand it well enough, are then averaged to find the value of the white point. By my estimate, there are not more than 500 almost white Mars pixels, so less bright pixels (and probably some almost black) are included in calculating the average and end up with the white point of the image being computed as 72.9%, I think.
You could experiment with Auto Range, where the thresholds are user-adjustable, and see whether reducing the white point threshold would drastically increase the white point value of your image.