I just read in Phillips led datasheet that led has 200.000 switching cycles. Would that count PWM cycles too? Because it would mean only few minutes of operation.
Also - is it ok for all DC led bulbs to be controlled by pwm?
PWM brightness control enables dimming of white LEDs without significantly changing their color temperature or CRI. It is however less efficient, and if done at lower frequencies, it can cause eye strain. PWM is also unsuitable for many video or photography applications. Current adjustment is more efficient and does not cause eye strain, but is more complex to implement and can cause the color to change at low brightness when used with white diodes.
Individual LEDs have an essentially unlimited number of switching cycles, so PWM itself will not reduce their lifespan. PWM drives LEDs less efficiently, so they will run somewhat hotter, which could negatively impact their lifespan, although if you are dimming them anyway they will likely be running well below dangerous temperatures.
Depending on your application, you may consider PWM good or bad.
The 200000 cycles for a bulb likely has the implication of a thermal cycle per on off cycle, meaning the components heat up and cool down to increase wear. Good LED bulbs will also use a constant current driver rather than just a rectifier for better safety and efficiency, although you haven't linked a datasheet, my guess is Philips has the money to pay an engineer for a decent design. If not abused and not strongly thermally cycled, the PWM cycle life of LEDs is extremely long, although I'm not qualified to guess how many orders of magnitude more than 200000 cycles you get, From experience, it's definitely orders of magnitude more. I find 2kHz manageable for PWM dimming, giving 120,000 cycles per minute, 7,200,000 per hour and 172,800,000 per day and my LEDs show no wear whatsoever after several days. Depending how the driver in the 200,000 cycle bulbs is built it could be a variety of components that fail first, and regardless of what actually breaks, my guess at the cause if they're concerned about on/off cycles would be wide thermal ranges caused by the unfavorable form factor of the bulb and possibly thermally insulating fixtures. They may be tested to 200,000 cycles based on some worst case condition, like an insulated pot light and may last longer in favorable conditions like an open air sconce.
This next part would fit better in user1850479's answer, but just an example taken from the datasheet of my favorite LED, CREE XP-G3 (5000K, S5). If all you care about is how much light you get per Watt, at 85\$^\circ\$C, you get 172 lumens at 350mA, but if we go further down the datasheet...
This graph shows voltage and current for the rated current of 350mA and a minimal current of 125mA. I added an example at max continuous current of 2A.
From it we can extrapolate that at 3.06V and 2A, the LED uses 6.12W.
At 350mA and 2.75V, the LED uses 0.9625W for a luminous efficacy of 178.7 lumens per watt.
If the LED is run at a minimal 125mA, you can see it will be at 2.64V, using 0.33W. I didn't bother getting too close to it's rated cutoff current. If we look to the next graph:
At 2A, output is 428% of rated, giving 736.2 lumens at a luminous efficacy of 120.3 lumens per watt, only 67% as efficient as at rated current, and requiring a huge cooling apparatus.
We can see (or rather I counted pixels and measured) that at 125mA, output is 37.5% of nominal, so 64.5 lumens, giving a luminous efficacy of 195.5 lumens per watt, 9.3% higher. This is similar to the gain from purchasing a better bin of LED or newer generation (XP-G3 vs XP-G2) or the efficiency gain you get from keeping the LED at 25\$^\circ\$C rather than 85\$^\circ\$C. The disadvantage is you buy 2.7 times the number of LEDs to start, but tripling a purchase order may reduce cost to mitigate, so you cost will likely be less than 2.7 times. This is also based on running the LEDs at 85\$^\circ\$C, so it serves as a conserverative estimate, since running the LED at 34.3% power will make it much cooler as the power density will be spread out and there will likely be more heatsink available though less is needed per lumen.
As far as the original question of whether PWM control is inherently bad, your gains in using constant current dimming below the rated current of this particular LED are moderate. If you're willing to hand tune each board closer to the minimum stable current, you can squeeze another 3-4%. In a high density power application like a flashlight where the LED is being run at it's maximum current and thermally limited, designing the driver to PWM the maximum current to produce lower settings would be foolish, with 33% of your power at stake.
In some applications where a linear visible response to duty cycle is desirable and full output power must be matched between multiple LEDs, as in RGB dimming, the simplicity and resulting low cost of PWM can outweigh the loss in efficiency.
Say you had a high power RGB LED display like a giant daytime visible screen on a skyscraper where you want large, powerful individual pixels and cooling such a large array is a complication, it might justify the engineering cost of using 4 LEDs per pixel instead of one and giving each pixel a drive circuit capable of calibration to current control and 0-100% dimming (microcontroller with a lookup table would be an easy way for a hobbyist). The gains would be increases in efficiency, LED longevity, and you could experiment to see how many pixels can be controlled at a given framerate by a single microcontroller. For instance if you wanted a shift register scheme similar to common RGB LEDs, One pixel requires up to 4 PWM signals worth of bits at whatever resolution you need (RGBW), so a constant current dimmer driver needs to be able to convert that to a duty cycle for the voltage converter that drives the constant current for each color of each pixel, and a separate duty cycle to switch the output all the way on and off at a lower PWM frequency for dimming below the cutoff threshold. This is complicated and a ton of work and... If you're willing to spend that much money anyway, you can have each pixel composed of series sets of optimally underpowered LEDs as dense as you can pack them if each pixel is 5 cm by 5 cm. This lets you use 0-100% PWM dimming the simple way, optimise LED life time and get the optimal efficiency current control dimming would have given you at all brightness levels. As you can see above, using 3 or 4 LEDs instead of one will get you optimally close to the cutoff current, and using a larger number may be worthwhile to spread heat better and give a more diffuse light source (leading to efficiency gains from less diffusion filtering being necessary) and increasing peak output capability and contrast as well as lifetime. Only when your power density requires more brightness than a full sheet of LEDs can produce will you have to go below max efficiency and you can deviate a fairly long way before losses become great.
Depending on your situation, PWM could cause of an unnecessary (and possibly large) efficiency loss or an increase in convenience and driver simplicity, and which is dominant will depend on design constraints like space and whether you can afford many more LEDs.