Yeah, I'm sure that I was flat out wrong now. I've sat down and done some calculations. I used the initial string tension and bow draw weight figures that DC supplied from his bow to make sure I was using real data. Then I assumed some angles and did some force balancing and trignometry.
Sure enough, the tension on the string does actually go down as you draw the bow. I assumed that at full draw the angle between the string protruding from either side of your hand was 90 degrees for ease of calculation. But if you put the data into a spreadsheet and vary that angle, you still have reduced string tension. If you increase that angle (which would correspond to letting off the draw a bit), and lessen the draw weight accordingly, the string tension will increase until it reaches the original max tension at brace.
This is totally counter-intuitive at first glance, but the math confirms it. What we have going on here is a pulley-like effect where your fingers are the pivot point. The tension on the limbs is no longer being held by 1 string, but by 2, so the tension on either length of the string will be smaller.
I would be curious to see if there was a point at which stacking (the exponential growth of draw weight near the end of the draw) would be able to force the string into a state of higher tension than at brace, or if the string angle would continue to decrease enough to outpace string tension until failure.