I'm not an expert on this subject, but as I understand it, it is possible for to convert 1080i to 720p very well. It requires very high end equipment.
However it is common to take shortcuts in this process. 1080i consists of a 540 line field for 1/60th of a second, then a second 540 line field in the next 1/60th, interlacing them to appear like 1080 lines over 1/30th of a second. 720p has 720 lines for 1/60th, updating the next 1/60th with another full 720 lines. So one can argue that along the vertical axis, that 720p conveys more information in 1/30th of a second than 1080i does.
So to convert to 720p, you take one of the 540 line frames and upscale it to 720 lines, which is altering the information. Very good, expensive scalers do this better than lower-end scalers, which might just double every third line to produce 4 lines out of 3, which equates to 720 out of 540.
It also has to downconvert the 1920 horizontal lines to 1280, losing 33% of the information.
What scares me about Dish going to 1280x720p across the board is that they just might use a lower end solution. And even in the best case, with the best equipment, it is still altering the information from the original.
By comparison, converting 1920x1080i to 1280x1080i throws away 33% of the original information, along the horizontal, just like converting from 1920x1080i to 1280x720p, however it doesn't alter the vertical content, thus it can be (& likely is) higher quality.
Native 1280x720p content is the output of cameras capturing the information at 1280x720, which has some advantages over 1920x1080i in capturing motion. All of the information captured is passed along to your TV. So it can look very good, as compared to a "cheap" downconversion to 720p which can look crappy due to the information loss.
Thus comparing true 720p vs downconverted 720p is not an apples to apples comparison. Downconverted CBS or VOOM may not look anywhere near as good as ESPN or FOX true 720p. One contains all of the original information, in its original format, the other is a combination of a loss of 33% of it's information along the horizontal with altered information along the vertical.
Now if any of the above is rife with error, I'll trust others to post more correct info.