So, what I did was find a field near the galactic plane with lots of bright stars (to have something to measure in a 50 ms exposure). I took a 10 second exposure, a 5 minute (guided, but not OT shifted) exposure, then 55 50 ms exposures, and then another 10 second exposure. I haven't got all the way through the analysis, but the initial results are interesting.
The two 10-second exposures had average PSF FWHMs of 0.430 and 0.512 arcseconds, which themselves average to 0.47 arcseconds. The 5-minute exposure has an average PSF FWHM of 0.452 arcseconds, pretty close to the 0.430 of the 10 second exposure taken immediately before it.
The 55 short exposures have the following distribution of FWHM values (I have only measured one star so far):
As a quantitative test of this, I took four frames, two with small FWHM (0.245 and 0.248) and two with large FWHM (0.675 and 0.567) values. The average of these is 0.434, close to the average of the entire set of 55. I then shifted these frames so that their brightest pixels were aligned. I averaged them and measured the PSF: 0.26 arcseconds. This is the ultimate promise of local guiding. How much of it can be gotten from coherent guiding is still to be determined.
UPDATE - I found a simple way to do this shift and add for the whole set of 55 frames. The resulting PSF FWHM is 0.33 arcseconds.
UPDATE 2 - Daniel correctly points out that, in real OT shifting, you are using the image from 50 ms ago to predict the shift now. I can't model that in my experiment, since each image took 30 seconds to read out. So, my experiment is a best case, and there may be an additional contribution to the degradation of the PSF.