Are they almost always kept confidential? Or is there protocol for sharing them?
In my experience, the contents of the comments of other referee reports are only made indirectly available. Since the authors are normally expected to provide a response to the reviews, the relevant criticisms and comments of the other referees are typically mentioned or discussed in that document. Outside of that, however, there's often little direct sharing of referee reports. None of the eight or nine journals for which I've reviewed (physics, chemistry, chemical engineering) have allowed me to see directly the reviews submitted by the other referees.
At any rate, the results are almost always kept confidential, unless it is an "open" referee process by design. (There are a few journals now that make the refereeing process a part of the publication record for a given paper; an example is The Cryosphere.)
Some journals/conferences have explicit guidelines that tell you to treat the reviews confidential. I'm not aware of any journal that makes the reviews and authors' response publically available when a paper is published.
I think publishing reviews for your papers would in general be frowned upon, even if there's no explicit rule saying that you can't. That said, I've been wondering about that myself and we had discussions about it at our school because the quality of some reviews is very bad and making them public might help improve the quality of peer reviewing in the long term.
I personally try to write reviews in a way that I wouldn't object to them being published with my name on it (although all the reviews I've done so far have been anonymous).
It varies substantially by field and journal. (As I keep saying, this is true of pretty much every question on this board.) Often they are shared not just with the author but the other reviewers, too. The author can show them to whomever they like. Beyond that, there's no formal sharing mechanism, and no real demand for one.
There are a few journals I've encountered where the written reviews are published alongside the paper itself (the BMJ's new open access journal comes to mind). Other than that, generally I've only seen them sent to the other reviewers (usually because they're BCC'd on the decision/review email sent from the journal editor) or kept entirely confidential.
Probably the most common thing that happens? The authors complaining about idiot reviewers to their colleagues.
In some subset of journals they are openly available. Who was the reviewer and what he/she wrote
BioMedCentral medical journals are often like that. See example here
you click on each article's publication history
In all journals I have encountered the reviews are kept by the journal through either an electronic submission/review system (digitally) or in paper copy (if they do not have such a system). In the electronic systems it is usually possible to view ones older reviews but not view other's.
The question whether these are confidential or not is far from clear in general. I would, however, venture to state that a review is treated as confidential unless explicitly stated otherwise, i.e. clearly stated to authors, editors and reviewers (remember that the review comments received may include overruling comments by the editor). Such is obviosuly the case with some journals with open discussion formats.
To some extent the confidentiality issue is new to me because in my field everyone pretty much assume reviews are. I have not heard of anyone trying to push for publicising a review (again, in my field). It would probably not go down well if it happened without consent from the parties concerned. It seems this is a sector that is largely unregulated other than in general terms and understanding.