What you are trying to do is complicated. Stabilisation of power supplies is not something that can be done easily over an internet forum. Here is what is happening... if you want the theory...
All control systems have limited bandwidth, so can be approximated as a low pass filter. A low pass filter has a phase lag (go through the maths: (1/jwC)/(1/jwC + R) and gives a negative phase shift which is -45 degrees at the 3dB point for a single order system.
So.... your control loop has probably got a phase lag that is pushing the response close to the point of oscillation. If your control loop gets near to 180 degrees of phase shift, you will notice peaking in the response and possible changes in the duty cycle. A capacitor across the top feedback resistor is pulling the control loop's response away from 180 degrees, back towards stability. This is why you should put a cap across the top feedback resistor and not the bottom one. A cap across the bottom resistor adds more phase lag and pushes you closer to instability. The top feedback cap is adding phase lead that counteracts the phase lag of the system's natural response.
Now, switched mode power supplies tend to be second order (they have an inductor and a capacitor), so they are more prone to instability. Linear regulators tend to be first order (only an output cap), so get nowhere near 180 degrees which explains why your linear solution is not oscillating and your switched solution is... I cannot be sure, but it sounds like this is the case.
Another theory is that you could be experiencing sub harmonic oscillation which is a characteristic of switched mode solutions. This is where the internal compensation of the inductor current sensing is not working properly and is causing alternate highs and lows in the peak inductor current. Increasing the inductor value can sometimes help in situations like this. If you want to read more on this, Google Slope Compensation and Subharmonic Oscillation.
To really get to the bottom of what is happening, you need to analyse the response of the control loop using a Bode Analyser. Google Middlebrook Loop Stability to find out more on this. RD Middlebrook wrote some papers on how to do this and it is common practice in the power supply world.
Or, it could be board layout. If you have a corrupt current sense signal, this will send the control loop crazy and you have no hope of rectifying the problem.
As I said, it is not easy.... but hopefully this has explained some of the theory behind what you might be seeing