We've been having some crystal oscillator problems wit a PIC16F946, (probably board leakage), and I've looked at the LP oscillator waveforms. I've got about 1.1Vpp at each end.(2.3Vpp across). Isn't that rather a lot for a typical watch crystal?
(The documentation for PIC16F946 is still 'preliminary' after 10 years, missing AC information, says that the driver circuit is new, and the logic diagram for the LP oscillator is clearly wrong, so I have a lack of confidence in the oscillator driver circuit.)