Reference annotation for mapping (human)?

[dvanic]

Question:
Which annotation is best for reference-based/guided mapping and assembly using Tophat/Cufflinks/other tools?

UCSC Comprehensive Gencode11 track, downloaded as a .gtf, has 2072303 records.

ENSEMBL66 [Homo_sapiens.GRCh37.66.gtf] has 2181086 records.

gencode11 (on which ENSEMBL is based(???)) has 2617472 records, of which:
1947538 gencode.v11.annotation.level_1_2.gtf
586834 gencode.v11.annotation.level_3.gtf
74062 gencode.v11.polyAs.gtf
9038 gencode.v11.2wayconspseudos.gtf

Or, what are the pros/cons of using each of them? And what is the actual difference?

Thanks in advance!

[dvanic]

*bumping up my own post, since this is still relevant*
(Thanks in advance!)