Corrected, 88% of the patents grant made by PTO that has been tested through an ex parte reexamination has been found to been incorrectly issued by the PTO.
The data is PTO's own yearly statistics, and is under the heading of issued certificate of reexamination, and one need to compare the period of 2011/09/30 -> 2012/09/30 to get the latest data. A certificate is the concluded decision after all the appeals are done. The statistics cover all types of patents, and all types of industries. Statistics for specific markets (say software) doesn't seem to exist.
The reason I don't bring up unchallenged patents, is that they are of indeterminate quality. The market values them less, and they shares many of the attributes of untested code. It might be good, it might be bad, but until the code is actually run and tested, nothing much can be said. Discussing the quality of untested code seems to me at least as exercise in futility. If we were to try asses the average quality of developed code, would we include untested code too, and if so, how?
On the positive side, patents that are used by entities that try to extract money from patent licensing do tend to get their patent contested. So the patents that are challenged, are also the patents that are in-use at patent suits and patent agreements.
