One one of our boards, we had intermittent USB failures (sometimes after 5 minutes, sometimes after 13 days of continuous operation) that were ultimately caused by slightly misaligned/misspaced USB D+/D- traces. We ended up having to force usage of USB full speed mode for USB high speed devices via a debug register (not unlike the article's case). Makes me feel slightly better that even people much more competent than I end up having to use such workarounds.
If I were to guess, given the tolerances in timing for USB 2.0, one line was slightly longer than the other on the board. That'd mean that the transition on D- would happen at a different time than the one on D+ causing corrupted data.
