Probably a number of factors Dave. A bad 'batch' would have been HUGE as they used this 6905 bearing in
many Honda motorcycles 2000 and on. I suspect perhaps more likely to
possibly be involved could be a lesser capable bearing manufacturer that they had a contract with for a number of years, concurrent with a poorly managed collar machining tolerancing.
I ordered a spare collar after my first flange bearing failure so that I could have a spare set of bearings on a collar already assembled, making tire changes quicker when/if I found bearings failed (my bike or others that come through my garage).
What surprised me was how different the Honda bearings fit the new collar as compared to the original collar that came in the bike. So during a tire change, I pulled the original collar and removed the bearings to measure various parts. My measurements showed the OD at the bearing seats to be about 0.001" different from each other on the two collars, and further collar OD's were different from where one bearing would seat vs. the other bearing on that same collar. I found the various Honda and aftermarket 6905 bearings I had on hand to have very repeatable ID.
So I put new Honda packaged bearings on the new collar and installed it in the ST given that the original collar allowed the bearings to fail in about 30kmi (I forget exact mileage). That setup (new collar, Honda bearings) didn't fail and was still with the bike when I sold it.
I didn't take the issue any farther than convincing myself that one Honda collar wasn't machined as appropriately as the other. One would need to see Honda's collar prints, specifications, and measuring a bunch of collars as produced to know what the issue really was.
But for designing proper bearing seats, 0.001" difference of two 'identical' parts is simply a glaring issue that does not lead to good, consistent bearing life. As a result one pair of 6905 bearings on a collar can last a long time, while another set of bearings on the very same collar will last less time just because the collar isn't optimum dimensions and slight bearing differences will become more prominent.
For perspective, on a recent high speed shaft-bearing/housing design I managed, I did the bearing calcs (they're more detailed than most realize). Following SKF's design criteria, I specified the bearing seats to x.xxxx" +0.0004" -0.0008". And that was for various bearings having 42mm bore and larger.
So my finding a 0.001" seat difference on a 25mm seat from the two collars I had in hand shows that Honda messed up [at least for one of the two collars I had in my hand].
Related... I see that the driven flange
inner bearing collar is also used in the rear drive assembly of the 2003+ VTX 1800 and VTX 1300 (as are the pair of 6905 bearings) ... I suspect those bikes also showed some premature driven flange bearing failures too for exactly the same reasons. Anyone know if that is true?