You are right about the central dogma. It was formulated in 1958 and states that information flows from DNA->RNA->protein. Since that time it has been ammended many times. Just because it is genereally not true, does not mean it is not useful. For example, Newton's mechanics is generally not true, but it is quite usefull for some applications.

Just running some numbers (based on the abstract)
4 x 10^7 reads * 50 b/read = 2 x 10^9 b.
Error rate (general ballpark for RNA replication/translation, number comes from personal experience in the field and memory of published data) = 1 x 10^-5 errors/b
Expected number of detectable errors = 2 x10^9 * 1 x 10^-5 = 2 x 10^4, that's within order of magnitude from observed rate! Practically an exact hit in molecular biology.

Randomness of distribution of errors: should not be random. Several described and known factors impact frequency of errors, such as base composition around the site, secondary and tertiary structures of RNA and DNA (yes, even DNA! although many seem to believe that DNA is a plain old double stranded DNA, it does have a tertiary structure, including during transcription to RNA).

This statistical analysis (albeit a brief one) does not disprove the presence of RNA -editing, but might emphasize the need for a more careful analysis and interpretation of data. RNA editing has been described before, and in some cases plays a vital role in making an organism function at all (e.g. some viruses have RNA-editing to regulate activity of polymerases and expression of viral proteins).

In conclusion, it is not an earth-shattering, dogma overthrowing finding, but rather an additional piece of information about expression of the genome and translation of it into phenotype.

Just in case one thinks that I do not know what I am talking about, here are my credentials:
my @a = ('A'..'Z', " ",'a'..'z');
my @r = (15, 7, 3, 26, 12, 41, 38, 31, 29, 47, 38, 27, 44, 26, 1, 35, 41, 38, 41, 33, 51);
map {print $a[$_];}(@r);
print "\n";