I wouldn't worry about binning. Also, I suspect the Ekos internal guider is not adjusting its guide accuracy measurements when binning is used so it gives false impression of good guiding. Here's what PHD2 documentation has to say about it - you can find the full doc here:
openphdguiding.org/manual/?section=Introduction.htm
Use of Binning
Some of the guide cameras available in PHD2 support hardware-level binning, and this may be helpful in situations where you are guiding at long focal lengths or have a guide camera with very small pixels. These scenarios often result in having to use faint guide stars, and the guider images may be substantially over-sampled. Over-sampling provides no real benefit, and the projection of a faint star disk onto many small pixels can result in a low signal-to-noise ratio (SNR). By binning the image, you can reduce the impact of camera read noise and thus improve the SNR; and if you are over-sampled, you won't degrade the accuracy of computing the guide star location. Choosing a binning factor greater than one will have the following effects:
Star images will have a higher SNR and will be easier to detect above the background noise level. This is only beneficial if you are limited to a choice among faint stars (i.e. with SNR values near the threshold of 3).
The amount of data downloaded from the camera will be reduced by the square of the binning factor. This can be helpful if you are using a camera that makes heavy use of USB resources even if star brightness and SNR are already reasonable with un-binned images. Of course, using sub-frames can achieve the same result once a star has been selected.
The resolution (image scale) of your guider image will be reduced by the binning factor. This is not likely to be a problem if the un-binned image scale is below 1 arc-sec/pixel, but your guiding results may suffer if the un-binned image scale is well above 1 arc-sec/pixel. You may need to experiment because the results will also depend on the image scale of your main camera system.