A SAM file will contain one line for each alignment. So if a read aligned to more than one position, it would show up multiple times in the SAM file.

BWA returns only one hit per read. In the SAM output, the XA field contains the alternative hits (format: (chr,pos,CIGAR,NM;)* )

I think it depends on your mapping tool and parameters. If your mapping tool select 1 random position when a read is mapped to different positions in reference genome, then you will have 1 entry (1 row for that read) in sam file. But if your mapping tool reports all different positions, then as Pierre said, you will get XA field (contains the alternative hits) in the same entry (no extra row) in sam file.

Also note that, if your read file contains 2 identical reads with different identifiers, then they will be reported in 2 different entries (rows) in sam file (this case is different from when the same read is mapped at multiple positions at reference genome).

Ah that's funny because with my colleages we just calculated something similar. Have a look to the ratio a/b calculated on your sam file:

a: samtools view [?] | sort | uniq | wc -l b: samtools view [?] | cut -f 1 | wc -l

some aligners return multiple hits on multiple lines. This is novoalign sam output using default parameters, for example:

W39CP:1373:2520 256 hsa-let-7a-1    6   1   23M *   0   0   TGAGGTAGTAGGTTGTATAGTTA >>>>>>=======9;;>===>== PG:Z:novoalign  AS:i:30 UQ:i:30 NM:i:1  MD:Z:22T0   CC:Z:hsa-let-7a-2   CP:i:5  ZS:Z:R  ZN:i:3  NH:i:3  HI:i:2  IH:i:3
W39CP:1373:2520 256 hsa-let-7a-2    5   1   23M *   0   0   TGAGGTAGTAGGTTGTATAGTTA >>>>>>=======9;;>===>== PG:Z:novoalign  AS:i:30 UQ:i:30 NM:i:1  MD:Z:22T0   ZS:Z:R  ZN:i:3  NH:i:3  HI:i:3  IH:i:3

This is great question and great answers. I have made my learning note to the SAM format, which might be helpful to others. Here is it:

http://onetipperday.blogspot.com/2012/07/deeply-understanding-sam-tags.html