USP tolerances in terms of %RSD (or %CV)
The most widely used and referred dissolution tolerances are based on the USP Acceptance Table. The results are evaluated in stages. This means repeats are allowed with relaxed tolerances and higher degree of variances for each subsequent test.
Stage 1: 
Test 6 tablets. Each unit not less than Q+5% dissolved. 
Stage 2: 
Test 12 tablets (including 6 from stage 1). Average is equal or greater than Q, but no unit less than Q15%. 
Stage 3: 
Test 24 tablets (including 12 from stage 1 and 2). Average equal and greater than Q, but no more than two units are less than Q15%, and no unit less than Q25%. 
(The Qvalues are provided in individual product monographs, representing expected percent drug release (dissolution) at times, such as 30, 45, 60 minutes etc.)
Considering the above criteria with a Qvalue of 80, one can obtain the following set for acceptable results.

1 
2 
3 
4 
5 
6 
Mean 
%RSD (CV) 
Stage 1 
95 
100 
91 
90 
98 
102 
96 
5 
Stage 2 
96 
88 
65 
110 
66 
65 
82 
23 
Stage 3 
55 
61 
92 
98 
105 
102 
85 
26 

103 
87 
77 
97 
93 
89 
91 
10 
Therefore, one may observe an RSD (CV) of 20% or more, and the results/product would be of acceptable quality for regulatory purposes. The expected high variability (RSD/CV) in results is built in the tolerances.
In this particular case, the test would meet the criteria at the stage 1, thus testing for the next stages would not be required. However, this may be a random phenomenon and one may get results as shown for Stage 2 or 3, at the first stage as well. These tolerances are for last dissolution sampling time, where dissolution is the highest with the least variable results. However, if dissolution results are to be reported for earlier times (such as for extendedrelease products), the expected variability would be higher.
The above discussion, therefore, clearly indicates that the dissolution results, in particular using Paddle/Basket apparatuses, are expected to be highly variable, often in excess of 20% RSD. Setting tolerances tighter, as often desired or suggested, would not be scientifically valid or achievable. In that case, one would face a higher number of failures without any apparent reason, as is the case with the testing of the USP PVT prednisone tablets where tolerances are usually set tighter than one would observe in real life, thus failures.