Finally there’s this unique case. Here’s a match in which the aces and doubles were not counted twice, and there’s only a minor discrepancy in Total Points Won between the ATP and another source, but there are nevertheless significant discrepancies (first-serve percentage, success on first and second serve).
Edberg-Courier, 1991 U.S. Open final
6-2, 6-4, 6-0
USA Today's boxscore:
1st serve percentage 56 55
Service winners 1 0
Aces 3 3
Double faults 1 4
Pct. 1st serve points won 84 59
Pct. 2nd serve points won 69 38
Service games held 12 6
Winners 33 16
Unforced errors 18 20
Pct. break point conversions 43 0
Pct. advances to net 72 55
Points won at net 68 12
Total points won 96 61
Time of match 2:02
The ATP has Courier serving 10 percentage points higher than this, and Edberg is up by 4 points. In success on first serve, Courier is down 4 points, Edberg down 2. Less significantly, they have Edberg at 2 aces and Courier winning 59 points.
The St. Petersburg Times and the Associated Press agree with USA Today that Edberg's success on first serve was 84%.
So this match (see also Becker-Pioline above) suggests that there may be significant mistakes in the ATP database that have nothing to do with counting aces and df's twice.