So, first of all, to get this out of the way, my methods of measuring 'performance' are not perfect, either using my eyeballs to watch whats going on through VRS, or using somebody else's code in the way of the statistics measurements that come from dump1090.
My setup comprises a co-linear antenna feeding a mini-circuits amplifier and then an EPCOS SAW filter into my airspy mini. I am located just north of London, so around this time of year I can expect to receive messages from 200-250 aircraft every minute throughout the day. Flightaware shows something like 800k-900k messages per day.
The 'single message tracks' comes from one of the field outputs from dump1090 stats. I do not know how this is calculated. However, it does correlate well with what I can see with my eyeballs in VRS. With my eyeballs I can see the following things happen: tracks are created that that have ICAO codes that look to be odd (looking up on several online databases they do not match any known active aircraft), they also tend to only ever get one or two messages and then disappear. This happened with the old airspy_adsb and -f 1 but it was not very often. With the new airspy_Adsb and -f 1 I can usually see 4/5 of these types of track almost constantly (and not the same ICAO code - just fairly random, indicating they are just bad decodes) my flightaware stats also show a massive jump in 'other' classified reports when I use the new version with -f 1. The other things that I can see hinting at bad decodes is things like call signs being given to the wrong ICAO code and mis-reported altitudes/speeds (last night there was a Robinson R44 doing ~450kts at 30kft nearby which I thought was a bit funny, turned out it was just in front of a descending airbus out to my Northern max range, looking back to the R44 - it was only a single message again - therefore leading me to believe it was simply an ICAO code decode error and was actually the airbus - I must admit I did not record the icao codes to compare if there was a single bit error possibility). All of this goes away the moment I set -f 0 and restart.
Going back though, I think what prog and I discussed earlier seems to be a plausible explanation for this.