1. The effest tools works on paired-end reads only as single ended data has not enough information about the sequenced fragments in order to estimate the insert size distribution and GC biases.
2. No, unaligned reads are not taken into account as we have no information about the transcript of origin.
3. I guess by "repetitive reads" you mean multimapping reads. There is no special treatment for multimapping reads, but you can further restrict the set of single isoform transcripts (through the -i flag) to the set of single isoform transcripts withouth a significant number of multimapping reads.

EDIT: 3'. In the case you mean PCR duplicates by repetitive reads: they are implicitly taken into account during the estimation of PCR biases, as they contribute the increased number of fragments in the GC bins with higher efficiency.

EDIT 2: One more thing regarding multimapping reads: effest filters the input alignments according to the minimum mapping quality specified by the -q flag (10 by default). If your aligner assigns a mapping quality lower than that to multimapping reads, than they are effectively discarded!