Do you have any idea of how high is the percentage of discarded transations under "normal" conditions?
I'm not sure what "normal" conditions are. If you mean no-one-trying-to-abuse-anything situation, then 0%. Not that every single modified txid would be an attempt to abuse, but there doesn't seem to be any good reason for doing it and certainly none of the major clients support it.
Then there is a reason why such collisions should be considered an "incident" and logged separately (for forensic purposes), especially if the volume is not that big as to make it impractical. Pity it wasn't being already done on a regular basis, at least on some "strategical" points of the network.
Would it be feasible for some miners to be logging those "incidents"?
Yes. In fact, you wouldn't even need to be a miner. Anyone relaying transactions (e.g. running the full bitcoin client) could, in theory, log these "incidents".
Someone actually did this to an extent after mtgox made their initial announcement. From the results he posted to this forum, probably only "counts" were logged (not each individual transaction). Their count was generally quite low, with a few spikes here and there (and a 24 hour break on Chinese New Year, if I recall correctly).
Again a pity it wasn't being logged in full even after the mtgox announcement. But anyways, that information could imply that the "attack" was not so generalised as to justify the ammount of damage mtgox wants everyone to believe. Maybe it even was a smoke screen to give some credibility to the announcement.
Also, being this type of attack some sort of race condition (combined with mtgox negligencies), the volume should have probably been huge in comparison to the sucessful "exploits", and this doesnt seem to have been the case.