5i25 configuration with only a G540, no daughterboards

I would use at least 5000 for all times to start

That seems to have helped a bit. It's far less likely to not change direction, however, it still drops the change every so often. Might an even larger number help that?

Might also be the step polarity
In general large values for the direction setup and hold times are
not harmful since direction is only changed at close to 0 velocity

I gave the direction signals a polarity change, without any improvement. As I use the 'tune' facility and jog an axis in each direction, there appears to be a hesitation (sounds like a bit of grumbling!) prior to an attempted change in direction by the motor. The actual direction isn't always changing ... perhaps 50% of the time.

might be a pncconf issue, I would try LinuxCNC

I understand connection between stepgen and nb. of chanel. What about inputs like E-Stop, and Limits. Are they also connected to specific channel or is it IO port nb in PnCConf?