Hi,
thanks to all for replies. I know the 8052.com is very valuable and your replies too. We we did not experiment during weekend and I was also out of WEB. So, below are some our conclusions:

Because the LM2937 was unstable after connected 100n cap (this is against the logic explanation), we decided to test the LM2937 on experimental board and under standard temperature condition. The ESR of 220uF cap temperature dependence we simulated by connecting of serial resistor.

Results of measuring:
1. LM2937 was blocked by 220uF + 0 ohm resistor: it is possible to connect next cer caps without the oscillations
2. LM2937 was blocked by 220uF + 2 ohm resistor: after connection of the cer cap from 100 to 220nF, the oscillations was appeared
3. LM2937 was blocked by 220uF + 8 ohm resistor: after connection of the cer cap above 10nF connection the oscillations was appeared

Conclusion: The stab LM2937 blocked by ellyt with ESR slightly under 3 ohms is going to oscillate after connecting next 100n capacitor.
On the real board are app. 30pcs of 100nF blocking caps connected after chokes and 2pcs of caps directly on output of LM2937. This the same situation like on experimental board. The oscillating was not appeared neither under -25°C when mentioned 2pcs of caps was disconnected.

Notes:
- According of Philips origin ellyt caps with low ESR ( 220uF, 16V) are following temperature dependences:
on + 20°C....................ESR value (app. 0,6 ohm)
on - 25°C....................ESR value is app. 5 times higher, then on +20°C
on - 40°C ....................ESR value is app. 3 times higher, then on -25°C
- ESR slightly depends on frequency too. ESR is increasing on lower frequencies.
- We pressume that caps from another producers has similar ESR dependences

Regards,
Vaclav