I think that pindel is fine. The issue may relate to the general issues of calling indels from NGS data.
Some things to consider:
Have you visually look at how the reads have aligned over the regions
in which indels were not called? They may be problematic regions,
like for example homopolymeric regions. I would load your aligned
BAMs into IGV
Check your mapping quality (MAPQ) filter thresholds (reads with
indels will automatically receive lo MAPQ unless we specifically
treat them differently
Consider your indel length in relation to your read length. As you're
using Ion Torrent, I guess that your reads are ~75bp? If you are
trying to detect indels from reads of this size, it will always be
Consider downloading your BAMs from the Torrent Server and then using
something like GATK IndelRealigner followed by HaplotypeCaller in
order to attempt to identify the indels.