As Jean-Karim pointed, e-values are not comparable between BLAST searches of two different databases. Bitscore won't help you either, because this score also depends on the size of the database.
Solution 1 (more accurate):
Instead of playing with BLAST scores by yourself, just use the InParanoid software, which finds orthologs and paralogs between two FASTA files (A and C). InParanoid is based on Reciprocal Best BLAST Hit, but it does a lot of things for you. As a result, you get a list of all orthologs and paralogs and, most importantly, confidence value for each ortologous/paralogous pair. In this way, you can apply t-test (if values are normally distributed) or Mann-Whitney test (if they are not).
Solution 2 (primitive, questionable):
If you don't want to use InParanoid, just simple RBH then you need to tell BLAST that the search space is equal among BLAST of the same query to multiple databases. Then, both, E-values and bitscores will be comparable among all results because the search space will be consistent. You do this by simply using
-dbsize flag when running BLAST locally. In my opinion, you can use the average of the sequences in A, B and C. For example, if A has 4000 sequences, B has 3000 and C has 5000 sequences, I would set -dbsize flag to 4000.
blastp -query A.fasta -db C.fasta -dbsize 4000