XRT observations were performed in what we call "Auto state", with the XRT picking the observing mode (Photon Counting or Windowed timing) depending on the count rate measured on-board.  This in general works great for GRBs or transients, it does not so well for sources like 1es1959 that sit at a countrate that is right at the switching point between PC and WT.
Two problems arise: pile-up in PC, and lost time due to mode-switching. Future observations will be taken in WT

I did a little investigation on the pile-up of this source and I concluded that the exclusion of the region of an inner circle with a radius of 10 pixels should be sufficient to avoid pile-up (that causes the hardening of the spectrum and an incorrect flux measurement), so I've extracted the PC spectra excluding that region.

For some observations the source was at a count rate high enough for the XRT to be in WT mode for all or a considerable amount of time.  For these observations I've also extracted the WT spectrum.  A first look at the spectral fits shows that for observations with data taken in both PC and WT the results are pretty similar, within 10% pretty much, so that's a good internal check.

For the spectral fits I've used a simple absorbed power law, with nH as a free parameter.
The best fits results are from spectra with 20 counts/bin, but I've also created qdp files with more binning to have fewer datapoints as you usually prefer.

So, I'm attaching these files:

Swift_XRT_SpectraPC.txt   ---> Spectra of observations in Photon Counting mode
Swift_XRT_SpectraWT.txt  ---> Spectra of observations in Windowed Timing mode
Swift_XRT_Spectra.txt  --> Merging of the two.