Perhaps it is beacause SMPS-s crap out switching noise, it can have components near MHz range, depending on design.

It can have components reaching far in the GHz range.
These signals are common mode and usually and won't be present between the power supply line and ground but may APPEAR to be present between those lines when an oscilloscope is connected.

i meant the dc bias. It is usually set by some voltage divider (or servo) and usually comes from one of the dc rails, isn't it so?

A lot of voltages in amps can be set by voltage dividers or current sources or reference voltages.
Mostly designers make an effort in decoupling those.
I know of very little designs (except in tube amps perhaps) where high resistance voltage dividers are used to 'bias' a circuit straight from the DC voltage rail and have no decoupling.
I call this a severe design error.

Then perhaps the ripple or noise on the dc rail can carry on to the input of the amplifier?

Perhaps in poorly designed amps yes... in well designed amps these voltages usually are decoupled or come from reference voltages.

Noise in MHz range passes through the smoothing caps and even filtering caps, even my nooby measurements show that, at least i believe that for now.

No they won't be actually present there, what you see on your scope are common mode noise voltages.
They will likely appear to be present in an equal amplitude all across the circuit (of course it depends on where you measure in circuit under what conditions).
I ... and many other designers, engineers, service eng etc all have experienced similar 'measurements'.

There is a simple experiment made by EEVBlog showing how low pass filter fails in really high frequencies:

Yes they do simply because low pass filters are made with real life components where inductors eventually don't work any more and small capacitances will pass extermely high frequencies.
We are talking >> 1MHz

Assuming capacitively coupled common mode noise doesn't show up in the output, but it can still interupt transistor (or tube) switching, can't it?
That is what really interests me.

Yes it can, usually by 'AM-detection' which can create a lot of problems.
It could potentially F-up fast input or intermediate stages and not pass slower output stages.
Usually designers simply take countermeasures by placing small value caps and having an input filter in place which attenuates diff mode HF signals so they can't cause problems.
These problems are common for semiconductor circuits more so than tubes.

I'm thinking it will not show easily on scope or can be easily heard, just that it is there on a really miniature scale and not perfectly unaudible. It should be one origin of nonlinear distortion.

Nice theory but ... I don't think so...
Non linear distortion is caused by the active parts in the circuit.
When it is unmeasurable it is also inaudible.
We can measure (with the right equipment) faaaaaaar below any audible limits.
Of course you can have a nasty oscillation at 100MHz and not see it with a 4MHz scope.
Doesn't mean it is there though.

As said before most RF garbage is mostly common mode and can only influence equipment if it is poorly designed.

Not comfortable sharing your views on SMPS for audio?
I think Sennheiser used it in their amplifier (by the looks of it), and it sounds pretty good, though definently not top shelf.

My views don't really matter that much as they are just my opinions based on what I value to be true.
Personally I have NO problems with using SMPS in audio equipment and no reservations about using linear power supplies.... as long as the circuits they feed are well designed and take these things into account.
PCB design becomes more important when using SMPS though.
This doesn't mean PCB/layout/grounding isn't important when designing stuff with linear power supplies, as chances are that linear stuff may well be connected
to other gear that is connected via SMPS OR SMPS stuff (TV, monitors, computers,DVD players, mediaplayers and whatnot) that is connected to the same mains.

If transistor needs to pass Mhz signal along with audio, isn't it just the kind of source for nonlinear distortion (influencing transistor work)?

Most transistors work well >100Mhz.
An audio circuit, however, doesn't really have to do anything >1MHz.
I would argue not even >100kHz so I see no reason to limit the BW of incoming signals above those frequencies and thus preventing amplification of those signals.
IF those signals are there anyway they come from poorly (or not) filtered DAC circuits such as NOS DAC's, NOT via mains through smoothing caps, regulators and local decoupling as these are common mode.
The amplitudes of these signals usually are very small to begin with.
The smaller a signal is the less it is affected by non-linearities of active components as for small signals the gain is quite linear.

I meant the kind of hard to measure why there is no hard proof (yet) why say Ragnarok is better amplifier than some (even very good) opamp based design. Also tube amps doesn't measure very good, etc.
It has been discussed elswhere, but measurement techniques for audio are lacking, i have gathered this is reasonably accepted view here.

Yes, it is a widely accepted view here by most pirates... but lets just say not by everyone...

I am convinced the discrepancy between subjective found 'sound' and measurements is NOT caused my insufficient measuring techniques but most pirates are convinced of quite the opposite, doesn't make either viewpoint more true.
That is when it comes to amplification and other aspects in the purely electrical/time domain.
As soon as electromechanical conversion is involved then measurement, recording and reproduction techniques are severely limited and incompetent.

A LOT of subjectively found 'good' tube amps indeed don't measure that well when compared to some other amps that may or may not be SS.
It is quite possible to make fairly wideband, very low distortion tube amps that measure well and sound well though, one doesn't exclude the other and it may measure well yet be perceived as less well sounding.
It is also quite possible to design well measuring and good sounding SS amps or well measuring yet perceived as bad sounding amps, here too one doesn't exclude the other.
Also quite possible to make SS amplifiers that sound great but measure poorly, here too, one doesn't exclude the other..

There is no obvious relation between measurements and perceived sound, but some measurements can indicate bad signal fidelity of waveforms, yet that poor fidelity, in an electrical sense, may not be perceived as bad sounding by every one, in fact it may even be the opposite (vinyl vs digital type of thing).
Of course so many people, so many opinions and so many views.

To stay on topic, if someone hears improvement on some power cable (without placebo effect) it is because flawed electronics design?

I don't think if someone hears improvements with power cables that in that case electronics are flawed.
All one can do is take the persons word for it that it wasn't placebo as there is never any 'electrical hard evidence' just opinions that may be shared by others.
Some are prepared to take the observants word for truth, others may not be inclined to do so.

To me the real question is HOW was the placebo/confirmation bias/perception effect eliminated from that 'test' and was that 'placebo-elimination technique' sufficient to completely rule out ALL of the possible 'traps' before blaming perceieved differences on the 'electric' side ?