I understand that they probably will be very expensive. But I have little idea of what kind would qualify. How much bandwidth? At what sampling rate? Any other parameters?
USB 3.0 runs the extra lanes at 5Gbps, which equates to a clock of 2.5GHz. So you will probably need at least 3GHz bandwidth, at an absolute minimum. Quite pricey! To see anything clearly, you'll want even more bandwidth as the signals have multiple harmonics - you would need at least the 3rd at 7.5GHz and preferably the 5th harmonic at 12.5GHz to see anything even remotely square.
But the thing that people forget is, a scope is only as good as the probes that it connects to. So not only do you need a scope with enough bandwidth, but also a probe which has the bandwidth too. The signals are also differential, so a differential probe is likely to be required.
At high frequencies like USB 3.0 runs at, electrical signals in wires are basically EM waves trapped in a waveguide (the cable). These signals are incredibly sensitive to impedance mismatches, so sticking any old probe with a long cable on it is just going to distort the signals.
You would need a probe which has very low capacitance and connected to the signals with as short of a run of extra cable as possible. Keeping the cable short essentially calls for an active differential probe. Expect such a probe to be in the region of $7k+ just on its own!
For every oscilloscope manufacturer you will find some rules of thumb, like "5 times the signal rate" or "You want to see at least the 5th harmonic" and it all really also depends on what you want to do with it. Do you just want to try decoding it? Do you want to judge the eye diagram quality with and without equalization?
As one of the examples, for signal quality examinations LeCroy recommends 2.5-3 times the bitrate (remember, they are kinda square waveish), which means for USB3 Super Speed which is at 4.8Gbps, an analogue bandwidth of 12-14.4 GHz which then translates to lots more sampling rate and really big chunks of money.
A corresponding LeCroy presentation can be found here
Now of course this is for measuring the signal quality, to see anything at all, you will probably get away with half the bandwidth.