W5200 Connection stability

Hello Eugeny.
We used tantalum capacitor 10uF at XTALVDD from beginning. Modules WIZ820io used to use ceramic capacitor 10uF + 100nF at this pin. But later they replaced 10uF ceramic with 10uF tantalum. This thread [url]Wiz820io stops responsing to PING after few hour of running] mentions about it. On bottom of page is document Improvement of WIZ820IO.pdf. And replacing ferrite bead with 0R rezistor suggest W5200 Chip Datasheet v1.4.0 on page 12. Yes, magjack is not powered in best possible way. Boards are soldered properly in reflow oven.
But I noticed something weird. It seems Your suspicion about oscillator was right. When I touched pins of quartz with finger it did not metter. But when I touched X1 with iron tweezer, link fell down and both leds went off. I tried it many times. Sometimes was connection recovered itself. Sometimes helped reset of board (hw reset is connected to wiznet too). And sometimes I had to disconnect board from power. It is exactly state described at the beginning.
So I connected oscilloscop probe to X0. When I “connected” tweezer and my body to X1, oscillator starts run at 75MHz. And sometimes it stayed at 75MHz even tweezer was “disconnected”. Unfortunately I could not check it on randomly frozen device. Any Idea?