This one has me a little puzzled.
If in the left screen clip you have CH2 on the oscillator out, and CH1 on a CPLD pin which is buffering the clock signal, then I'm not sure why removing the probe from the CPLD pin would make so much difference.
If you had both probes on the oscillator output it's easily understood with the extra capacitance as Martin mentions, but as you described it they are not so something else must be happening.
As far as the waveform not being square goes, it could be a few things:
Firstly, don't assume the oscillator output is particularly square, it's rise/fall times may take up a significant portion of each cycle. Check your datasheet to see what to expect. I just had a look at a typical 10MHz CMOS oscillator, and the rise/fall times were given as 10ns, so that's 20% of each cycle and will not look that square.
Here's a screen clip of a 10MHz waveform with 10ns rise/fall times, but with no high frequency roll off. You can see it's not far off what you are seeing in the right hand screen clip.
Secondly the probes can be set to 10x so they do not interfere with the oscillator output as much and the effects of capacitance are reduced, the ground lead should be as short as possible too, ideally you remove the ground lead and use a clip on the tip of the probe to the nearest ground on the PCB (e.g. oscillator ground pin or nearby)
I think a 10MHz (true) square wave shouldn't look too bad on the Rigol, I would expect it to look a lot better than the clips on the left anyway. Although the bandwidth is given as 50MHz, it will pass some higher frequencies as the roll off probably won't be so sharp. So you might get quite a bit of the 4th and 5th odd harmonics.
One way to get a reasonable idea of what to expect would be to feed e.g. a 100MHz sine wave of known amplitude and see what you get on screen.
For reference, I just took a clip (from this video) of a 50MHz Rigol displaying what I believe to be a 20MHz square wave:
EDIT - Those clips look a lot better. The reason the distance between the signal/probe ground point makes such a difference is the extra inductance added. This can cause the high frequencies to be attenuated. As mentioned above, ideally you would unclip the ground lead and use a small wire/clip instead to the signal ground point in order to make the loop as small as possible. Note that it looks like you may need to compensate your probe correctly - see the links below for tips (or your scope/probe manual)
If your scope has a probe calibration routine then run this.
Here are a few links on probing high speed signals:
Tips for Improved scope measurements
Probing High Speed Digital Designs
Probing High Speed Digital Circuitry
Couple of good books:
High Speed Digital Design: A Handbook of Black Magic - Johnson and Graham
High Speed Signal Propagation: Advanced Black Magic - Johnson and Graham