Yes, around 2000 apparent inversions a day on average. See the attached inversion log from February 14, 2003. The lines that start with /o are the start of an inversion and the lines that start with /# are the end. Last two columns are the timestamps of the bids and offers in HHMMSSMMM format, UTC+1 I think. Columns 6 and 7 are the best bids and offers, cols 5 and 8 are the next best. Cols 9 and 10 are the bid and offer exchanges with 0 being ISE, 1 is CBOE, etc. in order of fill likelihood. Cols 11 and 12 are bid and ask sizes. This is from the SIAC multicast consolidated options feed. Colimn 4 is the last quoting exchange, which may not match up to cols 9 and 10 due to batching within UDP packets.
As you can see from the file, some of the apparent inversions persisted for quite a while, even as late as Feb 2003. I don't have any of my runs from the period (and don't have time to forge any up in photoshop! -- sorry, I know it is the ET way, but I don't have the time for it just now) and it was pure luck that I even have this log file, but we did get some fills.
We would attempt execution of what looked to be the mispriced leg first, then if filled we hit the other leg, so we can't say for 100% certain that these were pure arb fills, since the fills were not simultanious. At that time, over FIX, even the fastest exchange (ISE) still took an average of over 1 second to fill.
so it is what it is