We use mascot for the peptide identification and use Scaffold for the protein identification. Scaffold, is a commercial implementation of peptide prophet and seems to do a very good job of clustering the bands together and identifying the right proteins.
Also there are a lot of annotations, graphical comparisons, etc included.
I must admit we haven't done a thorough comparison to what's out there in the last year because this works so well.
ONe thing we keep going back to testing is the Trans-Proteomic Pipeline (TPP). This open source tool is also a very good tool. We just prefer the little features of Scaffold a bit more.
It seems that the question does not receive much attention so I would like to pop it up a little. :)
As for our laboratory, we rely on X!Tandem for peptide identification and use utilites from TPP and OpenMS (http://www.openms.de) + custom Python scripts for the rest of the work.
The reason for using X!Tandem is not really deep. We just took what was the most popular open tool (and X!Tandem is quite popular according to the HUPO and ABRF studies). And we have not yet seen a study which clearly shows the superiority of one peptide ID tool over the others.
I've had great success with OMSSA. Once you get past the command-line interface it is pretty simple to use. It provides good results and is multi-threaded so it's relatively fast. I've never had an issue with its results files being too big, although by default it does output the top 30 hits per spectrum so that could be an explanation. If you have issues with it, post a question at http://www.sharedproteomics.com/forum/ and it will get answered quickly.
More is better... Several papers have shown that the algorithms ID overlapping, complementary populations of spectra. We use OMSSA, X!Tandem, X!Hunter, Comet and Myrimatch so far. You will ID more peptides and validate many peptides which are ID'ed by multiple algorithms. Then use IDPicker, MassSieve or Scaffold to parse and combine the outputs. We have an app which spins up MPI clusters on AWS to do this, proteomecluster.com (not open source, sorry). An open source alternative is PepArML. It adds on a machine learning function post-search to pull out more peptides. We have a version of MassSieve that can filter on PepArML outputs if you are interested.
We use Mascot, Sequest, X!Tandem (and rarely OMSSA) together, combining the results in Scaffold.
Why? We like Scaffold, so we support what Scaffold supports: Mascot is the golden standard, many seem to swear by Sequest, X!Tandem is free and easy to use, OMSSA was a bit tricky because of huge result files (Scaffold could not parse the binary output).